Techniques of Text Classification

Webology, Volume 12, Number 2, December, 2015
Home Table of Contents Titles & Subject Index Authors Index
Techniques for text classification: Literature review and current trends
Rajni Jindal
Department of Computer Science & Engineering, Delhi Technological University, Delhi, India.
E-mail: rajnijindal (at) dce.ac.in
Ruchika Malhotra
E-mail: ruchikamalhotra2004 (at) yahoo.com
Abha Jain
E-mail: me_abha (at) yahoo.com
Received October 10, 2015; Accepted December 26, 2015
Abstract
Automated classification of text into predefined categories has always been considered as a vital
method to manage and process a vast amount of documents in digital forms that are widespread
and continuously increasing. This kind of web information, popularly known as the
digital/electronic information is in the form of documents, conference material, publications,
journals, editorials, web pages, e-mail etc. People largely access information from these online
sources rather than being limited to archaic paper sources like books, magazines, newspapers etc.
But the main problem is that this enormous information lacks organization which makes it
difficult to manage. Text classification is recognized as one of the key techniques used for
organizing such kind of digital data. In this paper we have studied the existing work in the area
of text classification which will allow us to have a fair evaluation of the progress made in this
field till date. We have investigated the papers to the best of our knowledge and have tried to
summarize all existing information in a comprehensive and succinct manner. The studies have
been summarized in a tabular form according to the publication year considering numerous key
1 http://www.webology.org/2015/v12n2/a139.pdf
perspectives. The main emphasis is laid on various steps involved in text classification process
viz. document representation methods, feature selection methods, data mining methods and the
evaluation technique used by each study to carry out the results on a particular dataset.
Keywords
Machine learning; Text classification; Feature selection; Bag-of-words; Vector space model
Introduction
Text classification is the task of classifying a document under a predefined category. More
formally, if di is a document of the entire set of documents D and {c1,c2, c3,...,cn} is the set of
all the categories, then text classification assigns one category cj to a document di (Ikonomakis et
al., 2005). The documents depending upon their characteristics can be labeled for one class or for
more than one class. If a document is assigned to only one class, it is called single-label and if
the document is assigned to more than one class, it is called multi-label (Wang & Chiang,
2011). A single-label text classification problem can be further categorized into a binary
class problem if only one of the two classes is assigned to the document and this single-label
text classification problem becomes a multi-class problem if only N mutually exclusive
classes are assigned to the document. Text classification consists of document representation,
feature selection or feature transformation, application of data mining algorithm and finally an
evaluation of the applied data mining algorithm.
Now-a-days the amount of information available on the web is tremendous and increasing at an
exponential rate. Automatic text classification has always been an important application and
research topic since the inception of digital documents to manage the enormous amount of data
available on the web (Ikonomakis et al., 2005). It is based on machine learning techniques that
automatically build a classifier by learning the characteristics of the categories from a set of pre-
classified documents (Sebastiani, 2002). It plays an important role in information extraction and
summarization, text retrieval, and question- answering. Typically, most of the data for
classification is of heterogeneous nature collected from the web, through newsgroups, bulletin
boards, and broadcast or printed news scientific articles, news reports, movie reviews, and
advertisements. They are multi-source, and consequently have different formats, different
preferred vocabularies and often significantly different writing styles even for documents within
one genre. Therefore, automatic text classification is highly essential.
This paper provides an extensive study of the work which has been done till date in the area of
text classification highlighting the challenges which occur in classifying an unstructured web
content into a structured format. In other words, the paper aims to focus on elaborating the
dynamic and diversified nature of techniques available for classifying a given text into its pre-
defined categories and how these techniques have evolved over the past. This in turn will offer
new opportunities to the software practitioners and engineers working in this area. They can have
an in-depth knowledge about the progress made in the area of text classification beginning from
how the term text-classification has coined, followed by a summarization of the work done by
authors in this area and finally presenting open problems and issues for the researchers intended
to work in this area. After a thorough analysis, it was concluded that text classification is a
potential area of research and a lot of work can still be done towards improvising the existing
techniques and methodologies which have been used for classifying the unstructured text. The
paper presents a systematic review of previous text classification studies with a specific focus on
data mining methods, feature selection methods, the dataset and the evaluation technique used.
This review uses 132 text classification papers which will allow researchers to have a fair
evaluation of all the past studies and suggest possible new directions of research in different
areas concerned with text classification. The paper is organized as follows. Section 2 describes
the review process, in which we have defined our inclusion criteria and explained the selection
procedure. In this section we have also posed 7 research questions which will help us to collect
the necessary information. In section 3 we have classified the papers according to different
categories and have reported our review along with the important findings. Following this, we
have section 4 wherein we have reported the results using different graphical methods. Finally,
the review is concluded in section 5, in which we have also suggested some future directions.
Review Process
In this section, procedure used for selecting the relevant studies is discussed followed by an
inclusion/exclusion criterion. Then the research questions are highlighted which this review is
intended to answer.
Formulation of Research Questions
The most important objective of any review is to include maximum number of studies that are
filtered according to the defined inclusion criteria. Thus, selection of the relevant studies or to
have a suitable relevant subset of the papers is very essential (Malhotra & Jain, 2011). Following
two steps are undertaken to make the selection:
Step 1: This is the initial step in which we have searched various research related digital portals
such as ACM, IEEE, Springer, Elsevier, etc. Papers have been searched in various journals and
conference proceedings for appropriate selection (Sjoberg et al., 2005). There are a number of
important journals in which search has been done like Information processing and management,
Pattern Analysis and Applications, Information Retrieval, Knowledge Information System,
Pattern Recognition Letters, Journal of Intelligent Information Systems, Expert Systems with
Applications, Applied Soft Computing , Knowledge-Based Systems, Wuhan University Journal
of Natural Sciences , Information and Knowledge Systems, Neural Computing and Applications,
Machine Learning, Soft Computing, Decision Support Systems, Journal of Computer Science&
Technology, IJDAR, Information Sciences, Journal of Zhejiag University Science etc. All the
previous papers till date concerned with text classification have been collected and studied to
carry out an efficient review. This search was done by identifying the papers whose title or
abstract contains some of the relevant keywords such as text classification, text classification,
etc. Then, all the papers were scanned through and abstracts were read to identify the relevant
papers. This helped us to remove the irrelevant papers and obtain a smaller relevant subset.
Step 2: In this step, the subset of papers obtained in the first step was assessed for its actual
relevance. Final inclusion/exclusion decisions were made after retrieving the full texts. At this
step, we made the final decision or final subset as to which all papers should be included in this
review. The introduction and conclusion section of the papers selected in the initial stage were
read and hence a final decision was made. It is useful to maintain a list of excluded studies as
they are very useful in identifying the reason for exclusion. At the end of this step, we found 132
relevant studies related to our area of text classification.
Inclusion/ Exclusion criterion
Systematic reviews require explicit inclusion and exclusion criteria to assess each potential
primary study. The selection of primary studies is governed by inclusion and exclusion criteria
which should be based on the research questions (Catal, 2011).We included the papers in our
review if the paper describes research on text classification. This review does not describe all the
text classification models and the techniques used to develop them in detail for practitioners. Our
aim is to classify the papers with respect to their years, datasets, different feature selection
techniques, data mining algorithms and an evaluation measure. We included the papers published
in various journals and conference proceedings of digital portals which are of high repute like
ACM, IEEE, Springer, Elsevier. We have excluded the papers which did not include
experimental results. We did not exclude the papers wherein a new data mining algorithm was
not proposed, but instead a new feature selection technique or some new evaluation measure was
proposed. In other words, we included all the papers which were related to the field of text
classification in some or the other way. Our exclusion did not take into account the publication
year of paper or methods which have been used.
Formulation of Research Questions
Formulation of research questions (RQs) is very important to carry out the research. The purpose
of research questions is to let the readers know what the review is intended to answer. These RQs
were selected in such a way so as to ensure that there is a total coverage of text classification
area. In this review paper, we have addressed the following issues related to the area of text
classification:
RQ1. Which is the most popular journal in this area?
RQ2. Which year shows the maximum publications?
RQ3. Which data mining methods are widely used?
RQ4. Has the usage of modern machine learning methods increased over traditional statistical
methods?
RQ5. Which feature selection methods are commonly used?
RQ6. Which dataset is commonly used?
RQ7. Which is commonly used document representation method?
Classification of Papers
A number of different approaches have been studied to aid the text classification process. The
aim of a classifier is to use a set of pre-classified documents to classify those that have not yet
been seen. Figure 1 gives the graphical representation of a text classification process. The five
major branches include document representation, feature selection, constructing a Vector Space
Model (VSM), application of a data mining method and finally an evaluation of the text
classifier.
Figure 1. Text Classification Process
Document Representation
Document representation is the task of representing a given document in a form which is suitable
for data mining system i.e. in the form of instances with a fixed number of attributes. There are
several ways in which the conversion of documents from plain text to instances with a fixed
number of attributes in a training set can be carried out. Bag-Of-Words (BOW) is the most
commonly used word-based representation method. With this representation a document is
considered to be simply a collection of words which occur in it at least once. With this approach,
it is possible to have tens of thousands of words occurring in a fairly small set of documents.
Many of them are not important for the learning task and their usage can substantially degrade
performance. It is imperative to reduce the size of the feature space. One widely used approach is
to use a list of common words that are likely to be useless for classification, known as stop
words, and remove all occurrences of these words before creating BOW representation. Another
very important way to reduce the number of words is to use stemming which removes words
with the same stem and keeps the stem as the feature. For example, the words train, training,
trainer and trains can be replaced with train.
Feature Selection or Feature Transformation
Even after removing stop words from a document and replacing each remaining word by its
stem, the number of words in a BOW representation is still very large. Therefore, feature
selection method is applied to further reduce the dimensionality of the feature set by removing
the irrelevant words. It has a number of advantages like smaller dataset size, considerable
shrinking of the search space and lesser computational requirements. The goal is the reduction of
the curse of dimensionality to yield improved classification accuracy and reduce over fitting.
Methods for feature subset selection for text document classification task use an evaluation
function that is applied to a single word. Scoring of individual words (Best Individual Features)
can be performed using some of the measures, for instance, Document Frequency (DF), Term
Frequency (TF), Mutual Information (MI), Information Gain (IG), Odds Ratio (OR), CHI-
square statistic (CHI) and Term Strength (TS). All of these feature-scoring methods rank the
features by their independently determined scores, and then select the top scoring features.
Another technique to reduce the size of the feature space is referred to as feature transformation.
It is also known as feature extraction. This approach does not weight terms in order to discard the
lower weighted like feature selection, but compacts the vocabulary based on feature
concurrencies. Principal Component Analysis (PCA) is a popularly used method for feature
transformation. Some of the well-known feature selection metrics have been summarized in Table 1.
Constructing a Vector Space Model
Once a series of preprocessing tasks have been done (removal of stop words, stemming) and
relevant features have been extracted using a particular feature selection method, we will have
the total number of features as N which can be represented in some arbitrary order as t1, t2, ...,
tN. The ith document is then represented as an ordered set of N values, called an N-dimensional
vector which is written as (Xi1, Xi2, ..., XiN) where Xijis a weight measuring the importance of
the jth term tjin the ith document. The complete set of vectors for all documents under
consideration is called a VSM. There are various methods which can be used for weighting the
terms. The most popular method used for calculating the weights is called TFIDF, which stands
for Term Frequency Inverse Document Frequency. This combines term frequency with a measure
of the rarity of a term in the complete set of documents and has been reported to be the most
efficient of all the methods. Now, before we use the set of N-dimensional vectors, we will first
need to normalize the values of the weights. It has been observed that normalizing the feature
vectors before submitting them to the learning algorithm is the most necessary and important
condition. Comparative analysis of different document representation methods has been provided
in Table 2.
Application of a data mining algorithm
After feature selection and transformation, the documents can easily be represented in a form
that can be used by a data mining method. A data mining method can either be based on
statistical approaches known as the statistical method or can be a machine learning method based
on various supervised and un-supervised techniques of machine learning. There are many text
classifiers using machine learning techniques like decision trees (DT), naive-bayes (NB), rule
induction, neural networks (NN), K- nearest neighbors (KNN), and support vector machines
(SVM). They differ in their architecture and the approach adopted. Some of the well- known data
mining methods has been summarized in Table 3.
Evaluation of a text classifier
An evaluation measure is used to measure the performance of a text classifier. For each category
Ckwe can construct a confusion matrix as shown in the Figure 2 where a denotes the number of
true positive classifications, b denotes the number of false positive classifications, c denotes the
number of false negative classifications and d denotes the number of true negative
classifications. For a perfect classifier b and c would both be zero.
Predicted Class
Ck Not Ck
Actual
Ck a c
Class Not Ck b d
Figure 2: Confusion matrix for Category Ck
The value (a+d)/(a+b+c+d) gives the predictive accuracy. However, the standard performance
measures for text classification are recall and precision. Recall is defined as a/(a + c), i.e. the
proportion of documents in category Ck that are correctly predicted. Precision is defined as a/(a +
b), i.e. the proportion of documents that are predicted as being in category Ck that are actually in
that category. Each level of recall is associated with a level of precision. In general, higher the
recall, lower the precision, and vice versa (Yang &Pedersen, 1997). The point at which recall
equals precision is the break-even point (BEP), which is often used as a single summarizing
measure for comparing results. There are instances where a real BEP does not exist. It is
common practice to combine Recall and Precision into a single measure of performance called
the F1 Score, which is defined by the formula F1 = 2PrecisionRecall/(Precision+Recall). This
the product of precision and recall divided by their average which serves as yet another useful
measure used for evaluating the effectiveness of classifiers. These scores are computed for the
binary decisions on each individual category first and then averaged. When dealing with multiple
classes there are two possible ways of averaging these measures, namely, macro-average and
micro-average (Antonie & Zaiane, 2002). In the macro-averaging, one confusion matrix per class
is used; the performance measures are computed on each of them and then averaged. In micro-
averaging only one contingency table is used for all the classes, an average of all the classes is
computed for each cell and the performance measures are obtained therein. The macro-average
measure weights all the classes, regardless of how many documents belong to it. The micro-
average measure weights all the documents, thus favoring performance on common classes.
Table 1. Comparative analysis of different feature selection methods
S. Paper Technique Conclusion/ Advantage

No.
1 Tasci and Gungor IG , DF, Accuracy2, Local policy on IG, DF and Accuracy2 outperformed when the number of
(2008) AKS keywords is low and global policy outperformed as the number of keywords
increases, AKS selected different number of keywords for different classes and
improved the performance in skew datasets.
2 Tasci and Gungor LDA (Latent Models and discovers the underlying topic structures of textual data, IG
(2009) Dirichlet Allocation) performed best at all keyword numbers while the LDA-based metrics performed
similar to CHI and DF
3 Wang et al. (2012) LDA,IG Combines statistical and semantic information by building SFT, thus improving
the accuracy of short text classification
4 Yang and Pedersen DF,IG,MI, CHI, TS IG & CHI are most effective in aggressive term removal, DF has 90% term
(1997) removal capability and TS has 50-60% capability, MI has inferior performance
due to a bias favouring rare terms and a strong sensitivity to probability
estimation errors, DF, IG & CHI scores of a term are strongly correlated,
thereby meaning that DF thresholding is not an adhoc approach but reliable
measure
5 Zhen et al. (2011) Kullback-Leibler Measure differences of distributions between two categories and overcomes
(KL) divergence following disadvantages of CHI:-
based global feature CHI computes local scores of the term over each category and then takes
evaluation criterion maximum or average value of these scores as the global term-goodness
criterion. Now there is no explicit explanation on how to choose maximum or
average, Secondly, CHI cannot reflect the degree of scatter of a term
6 Bakus and Variant of MI Finds optimal value of redundancy parameter, outperformed IG, CHI, OR, CFS
Kamel(2006) (MIFS-C) (Co-relation based feature selection) and Markov blanket
7 Azam and Yao TF,DF Superior for smaller feature sets, have larger scatter of features among the
(2012) classes, accumulate information in data at a faster rate.
8 Yang et al. (2012) CMFS Measured the significance of a term in both inter-category and intra-category
with NB and SVM as the classifiers, superior to DIA, IG, CHI, DF, OCFS when
NB was used and superior to DIA, IG, DF, OCFS when SVM was used
9 Liu & Hu (2007) ARM Viewed a sentence rather than a document as a transaction
10 Qiu et al. (2008) DF,TF, TF-IDF,CHI A two-stage feature selection algorithm consisting of local feature set
constructed using DF, TF, TFIDF and global feature set using CHI
11 Meng and Lin (2010) DF, MI, CHI, LSI Reduced number of dimensions drastically, introduced the semantic model to
overcome the problems existing in the VSM
12 Meng et al. (2011) FCD, LSI Reduced number of dimensions drastically, introduced the semantic model to
overcome the problems existing in the VSM
13 Zifeng et al. (2007) CLDA Selects features using LDA but does not transform high-dimensional feature
space into low-dimensional feature space, better than IG and CHI
14 Torkkola (2003) LDA Reduced the dimensionality without sacrificing accuracy, 5718 number of
features reduced to 12
15 Fragoudis et al. Best Terms (BT) Fast performance, increases classification accuracy of NB and SVM,
(2005) complexity of BT is linear with respect to number of training set documents and
is independent from both the vocabulary size and number of categories
16 Pinheiro et al . (2012) ALOFT Ensures that every document in the training set is represented by at least one
feature, performs better than the classical Variable Ranking
17 Liu et al. (2012) Improved AM Removes those ambiguous features which are not removed by AM
18 Nuntiyagul(2005) PKIP Used for item banks, short textual data
19 Ko et al. (2004) Novel algorithm Measures the importance of sentences using text summarization techniques,
shows difference between important and unimportant sentences, considers
features from more important sentences
20 Wilbur and NBMBM Offers no significant advantage over plain MBM, word burstiness is so strong
Kim(2009) that additional occurrences of a word adds no useful information
21 Chen et al. (2007) Entropy Label Transforms multi-label data to single-label data before applying feature
Assignment (ELA), selection algorithms to solve multi-label feature selection problem, integration
IG, CHI, OCFS of four transformation approaches viz. All Label Assignment (ALA), No Label
Assignment (NLA), Largest Label Assignment (LLA) and Smallest Label
Assignment (SLA)
Table 2. Comparative analysis of different document representation methods

S. Paper Technique Conclusion/Advantage
No
Purpose: Work based on stemming and weighting methods
1 Song et al. Text representation schemes viz. stop words Schemes are corpus-dependent, for Reuters indexing and
(2005) removal, word stemming, indexing, weighting normalizing are important, for 20-NewsGroup weighting
and normalization and normalizing are important, among the five factors,
normalization is the most important, removal of stop
words from vocabulary is not harmful, word stemming is
harmful on Reuters and helpful on 20 NewsGroup
2 Harrag et al. Stemming methods : Light, Root-Based & Used for Arabic text classification, dictionary-lookup
(2011) Dictionary-Lookup Stemming stemming is superior for ANN and light-stemming is
superior for SVM
3 Leopold TFIDF weighting scheme Has larger impact on SVM performance rather than kernel
(2002) function alone, no pre-processing and feature selection is
needed for SVM
4 Lan et.al tf.rf (based on discriminating power) Term weighting scheme has a larger impact on the
(2005) performance of SVM rather than the kernel function
5 Wu et al. Term weighting scheme (based on word More accurate than the original weighting methods, reduces
(2012) clustering) dimensionality
6 Altnay Different weighting schemes Ordering of terms according to their discriminative abilities
(2012) is dependent on the weighting scheme
Purpose: To handle class imbalance problem
7 Lu et al. TFRd redundancy based term weighting Based on posterior probability distribution, promotes
(2009) scheme precision-recall, reduces sensitiveness to number of
features
8 Chen et Semantic re-sampling methods based on Uses global semantic information, DECOM deals with
al.(2011) probabilistic topic models DECOM & class imbalance by generating new samples of rare classes,
DECODER DECODER smoothens the data by regenerating all samples
in each class for data sets with noisy samples & rare classes
9 Sun et al. Different re-sampling and term weighting SVM learns the best decision surface in most test cases, for
(2009) methods using SVM classifiers classification tasks involving high imbalance ratios it is
therefore more critical to find an appropriate threshold than
applying any of the re-sampling or weighting strategies
Purpose: To modify the conventional BOW/ VSM representation
10 Deng (2009) Singular Value Decomposition (SVD) Reduced dimensionality to a great extent , discovered
important semantic relationships between terms
11 Wang (2009) Thesaurus of concepts built from Wikipedia Included semantic relations (synonymy, hyponymy, and
associative relations) thus expanding BOW representation
12 Hassan et.al Wikitology Enhanced text categorization by adding background
(2011) knowledge to documents, better than other knowledge
bases like Word Net, Open Project Directory (OPD),
Wikipedia
13 Ozgur (2012) An algorithm based on the extension of BOW Extracted fewer but informative features using the concepts
of lexical dependencies and pruning
14 Yun et.al A two-level representation model (2RM) Represents syntactic information at first level and semantic
(2012) information at second level, better than VSM
15 Pu et al . Local Word Bags (LWB) Represented a document as a set of tf-idf vectors, considers
(2007) detailed local text information ignored by BOW model
16 Jo (2009) Neural Network Classifier (NTC) Encoded documents into string vectors instead of numerical
vectors, removed problems of huge dimensionality and
sparse distribution
17 Kehagias Word and sense based classifiers The use of senses for text representation does not result in
(2003) any significant categorization improvement
18 Zhang et Concept representation and Sub-topic Represents the documents using extracted multi-words,
al.(2008) representation. have larger impact on performance of SVM rather than
kernel, subtopic representation outperformed concept
representation, linear kernel outperformed non-linear kernel
Table 3. Comparative analysis of different data mining methods

S. Author Technique Conclusion/Advantage
No
1 Lim et al.(2006) PSVM Allowed for automatic tuning of the penalty coefficient parameter C and kernel
parameter via MCMC method
2 Kumar & Gopal PSVM, Fuzzy PSVM Maintains constant training time irrespective of the penalty parameter C and
(2010a) categories, Fuzzy PSVM showed improved generalization over PSVM
3 Kumar &Gopal OAA-SVM, OAA performed better than OAO for uni-label text classification, OAA is
(2010 b) OAO-SVM suitable for text corpuses with small number of categories whereas OAO is better
on text corpora with large number of categories
4 Lee et al. (2012 a) Euclidean-SVM Has low impact on the implementation of kernel function and soft margin
parameter C, thus retaining the classification accuracy of SVM classifier,
Euclidean distance function replaces the optimal separating hyper-plane as the
classification making function of the SVM, consumes a longer time and has lower
classification accuracy than conventional SVM as Euclidean distance calculation
which inherits the characteristic of nearest neighbor approach suffers from the
curse of dimensionality
5 Dai et al. (2008) CHI based Algorithm Solves the problem of fine-text-categorization characterized with many redundant
features, Outperformed SVM and C4.5 algorithms
Purpose: To solve the multi-label text classification problem
6 Wang & Chiang Multi-label classifier Sample set from high dimensional space was mapped into a lower dimensional,
(2011) documents were categorized into multiple classes, probability that a document
belongs to a class was predicted
7 Wang & Chiang OAA-FSVM, Create multi-margin hyperplanes used to distinguish positive class from negative
(2007) & (2009) OAO-FSVM class and then the weight of each data set can be set according to its class, thus
solving the Fuzzy data problem, out-performed OAA-SVM and OAO-SVM
methods in multi-class text categorization problem
8 Namburuet al. PLS Better than SVM for multiclass categorization, SVM is more suitable for binary
(2005) classification as for multiclass categorization SVM requires a voting scheme
based on the results of pair-wise classification
9 Zelaia et al. KNN algorithm Based on Bayesian voting and SVD, documents represented by 15,000 features in
(2011) the BOW form and by 11,000 in the Bag-of-Lemmas were simplified to 300
features, consequently saving space and time
10 Schapire et BoosTexter system Embodies four versions of boosting, combines many simple and moderately
al.(2000) inaccurate categorization rules into a single, highly accurate categorization rule
11 Esuli et al.(2008) TREEBOOST.MH It is exponentially cheaper to train and to test than ADABOOST.MH
12 Chen et al.(2004) Boosting algorithm Achieves better performance on multi-label Chinese text categorization tasks than
other methods viz. NB and Rocchio algorithm.
Purpose: Work based on linear classification methods
13 Zhang & Oles LLSF, LR, NB,SVM Share similarity by finding hyper-planes that separate a class of document vectors
(2001) from its compliment, NB is worse, LLSF performed very close to the state-of-art,
LR performed as well as SVM
14 Basu et al. (2002) ANN and SVM SVM preferable for short text documents, less complex than ANN because
parameter that constructs the hyper-plane is very small, ANN performs large
matrix calculations on matrices
15 Wang et al.(2006) Optimal SVM It outperformed many other conventional algorithms
16 Li et al. (2011) VPRSVMRKNN Combines strengths of both SVM and KNN, VPRSVM filters noisy data which
reduces impact on RKNN classifier
17 Mitra et al.(2007) LS-SVM Based on LSI coefficient with Gaussian radial basis function (GRBF), LS-SVM
outperforms KNN, NB, SVM and NN based system.
Purpose: To solve PU-oriented text classification problem
18 Peng et al.(2008) Algorithm based on 1- Improved 1-DNF obtained more negative data with a lower error rate than 1-
DNF DNF, PSOC (Particle Swarm Optimization Classifier) performed better than
weighted voting method
19 Shi et al. (2011) Semi-supervised Used positive and unlabeled data based on tolerance rough set and ensemble
algorithm learning, tolerance rough set theory extracted a set of negative examples. SVM,
Rocchio and NB algorithms were used as base classifiers to construct an
ensemble classifier, Outperformed algorithms like SEM (Spy EM) and PEBL
(Positive Example Based Learning)
20 Pan et al. (2012) DCEPU Used concept drift by constructing a validation set and dynamic weighting
scheme to assign weight to each base classier in the ensemble, weighting
scheme considers not only the local weight of each base classier, but also a
global weight of each classier
21 Cabrera et al. Semi-supervised Removes the problem of supervised learning technique i.e. need of a great
(2009) algorithm number of training instances to construct an accurate classifier, does automatic
extraction of unlabeled examples from the Web
22 Lee and Kageura A virtual document Enlarged positive training documents, made virtual documents by combining
(2007) technique relevant document pairs for a topic in the training set, not only preserved topic
but even improved topical representation by using relevant terms that were not
given importance in real documents
Purpose: Work based on Multi-lingual text classification
23 Lee et al. (2006) LSI (unsupervised), Both the methods are complimentary, a hybrid system to overcome the
SVM (supervised) disadvantages of both approaches is required to give better results
24 Lee & Yang LSI and SVM SOM-based (self-organizing maps) supervised technique is used, a hybrid system
(2009) (unsupervised) is required to overcome the disadvantages of both the approaches
Purpose: Work based on KNN algorithm
25 Wan et al.(2012) SVM-NN approach Incorporates SVM to training stage of KNN classification, has low impact on the
implementation of parameter K thus retaining classification accuracy of KNN,
suffers from high time consumption
26 Dong et al.(2012) kNN algorithm Based on eager learning, overcomes lazy learning of traditional kNN algorithm,
decreases high computational expense
27 Wang & Wang TFKNN based on SSR Searches exact k nearest neighbors quickly, ranks all child nodes according to
(2007) tree distances between their central points and the central point of their parent
reducing searching scope and similarity computing
28 Soucy et al.(2001) KNN Reaches impressive results using very few features
29 Guo et al.(2006) kNN model kNN model outperforms the kNN and Rocchio classifiers
30 Wu et al. (2008) k-NN and M3-k-NN Majority voting method performed best when M3-k-NN is used while linear
voting method performed when k-NN is used, Gaussian voting method performed
best for both k-NN & M3-k-NN, M3-k-NN used less k value than k-NN and spent
less time to complete prediction than k-NN
31 Lu & Bai (2010) Refined KNN Weight measurement is based on variance, needs more running time than
traditional KNN but is far better than traditional KNN
32 Zhan and Chen GC,CNN, SNN,RNN, GC had highest average generalization accuracy when compared with CNN,
(2010) ENN SNN, RNN, ENN, especially in the presence of uniform class noise
33 Jiang et al.(2012) Improved KNN Based on clustering algorithm, reduced text similarity computation, outperformed
KNN, NB and SVM classifiers
34 Haifeng (2010) Improved KNN Based on skew sort condition
35 Yang (1999) DT, NB, NN, kNN kNN, LLSF and NN had the best performance, All other learning algorithms
Rocchio, LLSF performed well except for a NB approach, BPNN has limitations such as slow
training speed and can be easily trapped into a local minimum
Purpose: Work based on ANN algorithm
36 Li et al. (2009) Revised BP algorithm Based on automatically constructed thesaurus, removed disadvantages of
conventional BPNN
37 Li et al. (2012) MRBP, LPEBP MRBP and LPEBP alleviated the problems of standard BPNN, Semantic relations
of terms were considered using a CBT and WN
38 Wang et al.(2009) MBPNN, LSA Alleviated the problem of traditional BPNN, LSA removed the problem of VSM
by including the semantic relations between the terms
39 Zheng et Framework (LSA+ The weights and a bias-variance trade-off was achieved by adding a
al.(2012) RELM) regularization term into feed-forward NN, Learnt faster than conventional
algorithms such as feed-forward NN or SVM
40 Harrag et SVD-based MLP/RBF MLP classifier outperformed the RBF classifier, SVD-supported NN classifier
al.(2010) was better than the basic NN for Arabic text categorization.
41 Ruiz and Feed-forward NN Hierarchical structure performed better than equivalent flat model, Used divide
Srinivasan (2002) and conquer principle, comparable performance with respect to the optimized
Rocchio algorithm
Purpose: Work based on Centroid-based classifier
42 Nguyen et al. CFC, CFC-KL, CFC-JS CFC leads to poor performance on class-imbalanced data, CFC prunes terms that
(2012) (JensenShannon) appeared across all classes discarding non-exclusive but useful terms, CFCKL
was generalized to handle multi-class data by replacing KL measure with multi-
class JS divergence (CFCJS), KL and JS weighted classier outperformed
baseline CFC and unweighted SVM
43 Tan et al. (2011) Model Adjustment Deals with model mist problem of centroid classier, uses training-set errors
(MA) algorithm and training-set margins in contrast to methods like Weight Adjustment, Voting,
Renement, Drag-Pushing and therefore cannot guarantee generalization
capability of base classiers for unseen examples, converges to optimal solution
for a linearly separable problem
Lo (2008) proposed a mechanism to facilitate website management, named as WebQC which

used P-control chart to control web service quality. It gave a warning signal if the complaining
rate is higher than the upper control limit. In the paper by Couto et al.(2006), a comparative
study of digital library citations and web links in the context of TC was presented. It was
concluded that measures based on co-citation are the best performers for the web directories and
bibliographic coupling measures are appropriate for digital library containing scientific papers.
The work by Saldarriaga et al. (2010) categorized online handwritten documents based on their
textual contents using KNN and SVM algorithms. The effect of word recognition errors on the
categorization performances was analyzed, by comparing the performances of a categorization
system with the texts obtained through online handwriting recognition and the same texts
available as ground truth. Paquet et al.(2012) proposed an approach to categorize handwritten
document which is based on the detection of some discriminative keywords prior to the use of tf-
idf representation. Results show that the discriminative keyword extraction system leads to better
recall/precision tradeoffs than the full recognition strategy. In the paper by Farhoodi & Yari
(2010), two efficient machine learning algorithms were examined for Persian text document.
Experiments showed that the performance of KNN is better than SVM for Persian text
classification. Lia and Mu (2010) proposed an incremental learning algorithm on large-scale
corpus for Chinese text classification. In this study, an approach based on SVMs for web text
mining of large-scale systems on GBODSS was developed to support enterprise decision
making. Experimental results showed that this approach has good classification accuracy by
incremental learning and it was seen that speed up of computation time was almost super linear.
In the paper by Zakzouk and Mathkour (2012), three binary text classifiers viz. SVM based on
evolutionary algorithm, C4.5 and NB were built to test the cricket class of SGSC. It was
observed that Nave-Bayesian leads the pack with best effectiveness ratios overall. Wermter
(2000) showed that neural network can be used for tasks like text routing. This was illustrated
using different architectures and different corpora. In the paper by Liang et al.(2006), a new
dictionary-based text classification approach is proposed to classify the chemical web pages
efficiently. After automatic segmentation on the documents to find dictionary terms for
document expansion, the approach adopts latent semantic indexing (LSI) to produce the final
document vectors, and the relevant categories are finally assigned to the test document by using
k-NN algorithm.
Results
74 journal papers and 58 conference proceedings have been evaluated in this review
systematically. Each subsection of this section will address the respective research question
listed in the above section.
Relevant text classification journals (RQ1)
We used papers on text classification in 24 journals and these journals with three or more papers
are displayed in Table 4, together with the corresponding number, proportion, and cumulative
proportions of papers (Catal, 2011). Proportions and cumulative proportions have been
calculated by considering only the number of journal papers in review. 7 journals shown in Table
4 include 68% of all journal papers in review.
Table 4. Most important text classification journals
Rank Journal Name # Papers Prop (%) Cumulative Prop (%)

1 Expert Systems with Applications 16 22 22
2 Information Processing &Mgmt 8 11 33
3 Information Retrieval 7 9 42
4 Knowledge Information System 7 9 51
5 Pattern Recognition Letters 5 7 58
6 Pattern Anal Application 4 5 63
7 Journal of Intelligent Information Systems 3 4 67
Year showing the maximum publications (RQ2)
Figure 3 is a curve which plots publication year on the x-axis and the number of papers
published in that year on the y-axis for papers in review. We have reviewed in total 132 papers
on text classification. Out of these, 44% of papers are conference proceedings and 56% of papers
are journal papers. As can be seen from the curve, 22 papers were published in the year 2012
which represents the maximum publication year in this area followed by the year 2011 which
shows 17 publications. We can also observe that majority of the papers have been published after
year 2004. Figure 4 is a clear indication of our observation. In this figure papers have been
classified into two groups: papers published before 2004 and papers published after year 2004. In
total, 22 papers have been published before year 2004 and 110 papers have been published after
year 2004. It is clearly seen that the popularity of text classification area increased drastically
after year 2004 and thus researchers should only examine papers published after year 2004 to
reach the most important papers.
Figure 3. Number of papers per year in review
Figure 4. Distribution of papers after year 2004 (Number of papers /percentage of total papers)
Widely used data mining methods (RQ3)
From an extensive literature survey done in the area of text classification, it was observed that
the frequently used data mining methods are SVM, KNN, NB, ANN, Rocchio algorithm and
Association rule mining (ARM). These methods machine learning algorithm along with the
number of papers using these methods is shown in Table 5. As it is clear from the table, SVM is
the most popular used by the researchers in their work. Many of the authors have worked on
SVM algorithm and proposed its advanced version to better enhance the applicability of this
algorithm, thereby improving the performance of text classification. KNN algorithm is the
second popular method used by the researchers as it is used in 31 papers. Similar to SVM, the
authors of these papers have also proposed different variants of KNN and then compared the
performance of their proposed KNN algorithm with the different machine learning algorithms to
show that new KNN algorithm performs better than conventional algorithms. Finally, we have
NB algorithm which is used in 23 papers and thus falling under rank 3. It can also be clearly seen
that 65% of the papers are using SVM and KNN algorithm for categorizing the text and only 35
% of the papers are using other methods apart from KNN and SVM algorithms. This distribution
is shown in the Figure 5. This clearly indicates that SVM and KNN algorithms are amongst the
most popular machine learning algorithms used by the researchers. Out of a total of 132 papers,
88 papers used SVM and KNN algorithms and 44 papers used other data mining methods.
Figure 5. Distribution of machine learning methods (Number of papers/ percentage of total papers)
Table 5. Most important data mining methods used
Rank Data Mining Methods # Papers

1 SVM 55
2 KNN 31
3 NB 23
4 ANN 10
5 Rocchio Algorithm 9
6 Association Rule Mining 4
Distribution of data mining methods (RQ4)
Distribution of data mining methods which have been used in the papers is shown in Figure 6.
Methods have been divided into four groups: statistical methods, machine learning based
methods, statistical methods + machine learning based methods and the unknown category. If
machine learning based methods are used together with statistical methods in the same model,
method of that paper is marked as statistical methods + machine learning based methods. Some
of the authors (Tao et.al, 2005; Altinacy&Erenel, 2010; Luo et.al, 2011) have not used any of the
data mining method in their paper as they have proposed a new term-weighting method and have
compared its performance with the existing term- weighting schemes. Method of that paper is
marked as unknown methods. It has been observed that 86% of papers used machine learning
based methods and only 6% of papers used statistical methods like Hidden Markov Model
(Frasconi et al., 2002), Logistic Regression (Zhang and Oles 2001; Yen et al. 2011), Partial Least
Squares (Namburu et al., 2005), Linear Least Square Fit (Yang &Pedersen, 1997; Yang, 1999;
Zhang &Oles, 2001). Because statistical methods are considered black-box solutions and these
models are highly dependent on data, it is promising to see that more researchers are exploring
the potential of machine learning methods to predict text classification modules. As shown in
figure 6, out of a total of 132 papers, 8 papers used statistical methods, 113 papers used machine
learning methods, and 8 papers applied statistical methods together with machine learning
methods.
Figure 6. Distribution of data mining methods (Number of papers/ percentage of total papers)
Widely used feature selection methods (RQ5)
Methods which are widely used by the researchers are displayed in Table 6 along with the
number of papers using these methods. We can conclude from the table that CHI is the most
widely used method followed by IG as the second most popular method. It can be seen from the
table that CHI, IG and MI are amongst the most popular feature selection methods used by the
researchers in their work. Many of the authors have worked on these methods and proposed their
advanced version to better enhance the applicability of the method, thereby improving the
performance of text classification. Also many of the authors have proposed their own feature
selection method considering these popularly used methods as the base of their theory.
Table 6. Most important feature selection methods used
Rank Feature Selection Methods # Papers

1 Chi-squared test(CHI) 21
2 Information Gain (IG) 20
3 Mutual Information(MI) 16
4 Latent Semantic Indexing (LSI), Singular Value Decomposition (SVD) 10
5 Document Frequency (DF) 8
6 Term Strength (TS) 3
7 Odds Ratio(OR) 3
8 Linear Discriminant Analysis (LDA) 3
Widely used datasets (RQ6)
There are a number of datasets which are used by the researchers to conduct the experiment in
order to evaluate the performance of the data mining method applied. It has been observed that
researchers have mainly used the datasets from the famous machine learning repository called
UCI (University of California Irvine) which consists of several public datasets. Table 7 shows
the three popularly used dataset. It is clear from the table that the dataset namely Reuters-21578
is the most widely used dataset which is collection of documents that appear on Reuters financial
newswire service. 20NG is the second most popular dataset which is a collection of 20,000
newsgroup documents, partitioned across 20 different newsgroups. Finally, we have Web KB
dataset which consists of WWW pages collected from computer science department of various
universities. There are various other kinds of datasets used by the researchers like the datasets
consisting of medical data, E-mail data, mathematics data etc.
Apart from this, few authors have also analyzed the software project reports available in different
open source software repositories for predicting various aspects of software engineering.
Menzies and Marcus (2008) have analyzed the defect reports available in the PITS database of
NASA and presented an automated method named SEVERIS which is used to assign the severity
level to each of the defect found during testing. Assigning the correct severity levels to defect
reports is very important as it directly impacts resource allocation and planning of subsequent
defect fixing activities. Runeson et al. (2007) and Wang et al. (2008) have also analyzed the
defect reports and developed a tool that would be used to detect duplicate reports using Natural
Language Processing (NLP). Cubranic and Murphy (2004) analyzed an incoming bug report and
proposed an automated method that would assist in bug triage to predict the developer that would
work on the bug based on the bug description. Canfora and Cerulo (2005) discussed how
software repositories can help developers in managing a new change request, either a bug or an
enhancement feature. Lucca et al. (2002) analyzed the maintenance requests coming from the
customers in the form of a ticket (containing the description of the request) and developed a
router that would work around the clock and would keep dispatching the maintenance requests to
an appropriate maintenance team. Huang et al. (2006) analyzed the Non-Functional
Requirements (NFRs) as specified by the stakeholders during the requirements gathering process
and developed an automated technique that is used to classify them on the basis of its type, thus
leading to the detection of NFRs early in the development life cycle.
Table 7. Most important dataset used
Rank Dataset Used # Papers

1 Reuters-21578 71
2 20-Newsgroup 36
3 Web KB 15
Distribution of document representation methods (RQ7)
Distribution of document representation methods which have been used in papers is shown in
Figure 7. Papers have been classified into two groups: papers using Vector Space Model (VSM)
as its document representation method and papers using some other method (apart from
VSM).We have done this classification because it was observed from the literature survey that
majority of the papers used VSM and only a few papers proposed a new method for document
representation. As it is clear from the figure, 83% of the papers used VSM for representing the
document and only 17 % of the papers used other methods.
This clearly shows that VSM is the most common document representation method used by the
researchers. Out of a total of 132 papers, 109 papers used VSM and only 23 papers used a
modified version of VSM for representing the document as Bag-Of Words. For instance, the
paper by Frasconi et al. (2002) used the BOW representation resulting from a multi-nominal
word event model using Hidden Markov Model (hmm) for classification. Kehagias et al. (2003)
used a word & sense- based method for representation. The work by Kim and Kim (2004) used
the concept of passage based document wherein the document is split into passages &
categorization is done for each passage & finally document categories are merged with passage
categories. An and Chen (2005) represented the document in a subspace of the dimensionality
using an algorithm based on concept learning. Doan (2005) proposed a document representation
method based on fuzzy set theory. Pu et al.(2007)introduced the concept of Local-Word-Bag and
Zhang et al.(2008) proposed multi-word document representation. Srinivas et al.(2008) proposed
a MFCC algorithm (Multi-type Features Co-selection for Clusters) for representing text
document as a projection on clusters formed from the input dataset. Few authors also used the
concept of Rough Set Theory for representing the document (Zhou and Zhang 2008; Shi 2011).
Many authors extended the traditional VSM representation by using Wikipedia as a thesaurus to
consider semantic relationships between key terms (Li et al., 2009; Wang et al., 2009; Li et al.,
2012; Yun et al., 2012). Jo (2009) encoded the documents into string vector (rather than numeric
vectors) to avoid the problems of huge dimensionality and sparse distribution which are inherent
in encoding documents into numerical vectors. Lee et al. (2012b) used the Bayesian
vectorization technique and Wu and Yang (2012) used term clustering algorithm for
representation.
Figure 7. Distribution of document representation methods (# papers/ percentage of total papers)
Conclusion
To retrieve specific information from web is like finding a proverbial needle in the haystack. In
this work, the needle is that single piece of information a user needs and the haystack is the large
data warehouse built up on the web over a long period of time. Text classification is emerging as
one of the most prominent technique to handle this problem. In this paper, we have reviewed the
text classification papers since 1997 to 2012 published in conference proceedings and journals of
high repute to evaluate the progress made in the area of text classification so far. This review
would help future research based on the past studies. We have evaluated the papers with a
specific focus on types of data mining methods, feature selection methods, the dataset and the
evaluation techniques used by each study to carry out the results. Following trends were
observed in this work:
o Large number of datasets was used to evaluate the results to provide more accurate and
generalized results. It was also observed that more number of public datasets was used for
text classification because repeatable, refutable and verifiable models can only be built
with public datasets. From the review, we can conclude that majority of the researchers
have made use of the datasets available in UCI repository which has a collection of wide
range of public datasets. As many as 122 papers out of a total of 132 papers have made
use of the three most popular datasets available in UCI repository viz. Reuters-21578, 20-
Newsgroup and Web KB. Some authors have also made use of private datasets which are
not freely available and therefore it is not possible to compare results of the studies using
private datasets with results of our own models. Thus, we should make use of public
datasets available in UCI repository.
o The review clearly indicates that the most common method for representing a document in
text classification is the Vector-Space-Model which represents each document as a vector
consisting of an array of words. It is seen from the review that 83% of the papers are using
VSM technique for representing the document and only 17% of the papers are using other
methods. Once the document is represented as Bag-Of-Words, we reduce of the
dimensionality of the dataset by removing features that are considered irrelevant for the
classification. There are a number of methods proposed in the literature for feature
selection, but Chi-squared statistic and Information Gain are considered as the most
widely used methods.
o As specified in this review, machine learning models have better features than statistical
methods. Therefore, we should increase the percentage usage of the models based on
machine learning techniques. It has been observed that 86% of papers used machine
learning based methods and only 6% of papers used statistical methods. Among the
various machine learning algorithms studied in the literature, it has been observed that
SVM and KNN algorithms are the most widely used machine learning algorithms in the
area of text classification. These algorithms have been used by 65% of the papers. While
some of the authors have also made use of the statistical methods, but their use has been
very limited over the past few years because these methods are black-box solutions and
are highly dependent on data. It is promising to see that there is a drastic shift from
traditional statistical methods to modern machine learning methods.
From the literature survey done so far, it was observed that very little work has been done in
analyzing the software project reports which play a very important role in improving software
quality. Software repositories consist of different kinds of project reports which when analyzed
using text classification techniques can help in assisting project managers and developers in their
SDLC activities. Data contained in software repositories have generated new opportunities in
various directions such as change propagation, fault analysis, software complexity, software
reuse and social networks.
For instance, defect descriptions of given software can be analyzed in order to predict the
severity of defects by developing a new and automated method which can assist the test engineer
in assigning severity levels to defect reports. Building an agent using text mining techniques
would lead to saving of resources like time, manpower and money as text mining and machine
learning methods are low cost, automatic and rapid. Similarly, maintenance requests (i.e. tickets)
for a large, distributed telecommunication system can also be analyzed in order to route them to
specialized maintenance teams by developing a router that would work around the clock and
would keep dispatching the maintenance requests coming from the customers in the form of a
ticket (containing the description of the request) to an appropriate maintenance team. The system
would be able to balance the workload between different maintenance teams and there would be
lowest misclassification error as routing is done without human intervention. Also, Non-
Functional Requirements (NFRs) as specified by the stakeholders during the requirements
gathering process can be analyzed and then classified on the basis of its type, thus leading to the
detection of NFRs early in the development life cycle.
References
Altinay, H., & Erenel, Z. (2010). Analytical evaluation of term weighting schemes for text
categorization. Pattern Recognition Letters, 31, 13101323.
Altinay, H., & Erenel, Z. (2012). Using the absolute difference of term occurrence probabilities in
binary text categorization. ApplIntell, 36, 148160.
Amine, B.M., & Mimoun, M. (2007).WordNet based cross-language text categorization. IEEE/ACS
International Conference on Computer Systems and Applications, AICCSA.
An, J., & Chen, Y.P.P. (2005).Keyword Extraction for Text Categorization. Proceedings of the IEEE
International Conference on Active Media Technology AMT.
Antonie, M.L., & Zai'ane, O.R. (2002). Text document categorization by term association.
Proceedings of the IEEE International Conference on Data Mining, ICDM.
Azam, N., & Yao, J.T. (2012). Comparison of term frequency and document frequency based feature
selection metrics in text categorization. Expert Systems with Applications, 39, 47604768.
Bakus, J., & Kamel, M.S. (2006). Higher order feature selection for text classification. Knowledge
Information System, 9(4), 468-491.
Basu, A., Watters, C., & Shepherd, M. (2002). Support vector machines for text categorization.
Proceedings of the 36th Hawaii IEEE International Conference on System, HICSS03.
Cabrera, R.G., Gomez, M.M.Y., Rosso, P., & Pineda, L.V. (2009). Using the Web as corpus for self-
training text categorization. Information Retrieval, 12, 400415.
Canfora, G., & Cerulo, L. (2005). How software repositories can help in resolving a new change
request. Workshop on Empirical Studies in Reverse Engineering.
Catal, C. (2011). Software fault prediction: A literature review and current trends. Expert Systems with
Applications, 38, 4626-4636.
Chang, Y.C., Chen, S.M., & Liau, C.J. (2008). Multi-label text categorization based on a new linear
classifier learning method and a category-sensitive refinement method. Expert Systems with
Applications, 34, 19481953.
Chen, E., Lin, Y., Xiong, H., Luo, Q., & Ma, H. (2011). Exploiting probabilistic topic models to
improve text categorization under class imbalance. Information Processing and Management,
47, 202-214.
Chen, J., Zhou, X., & Wu, Z. (2004). A multi-label Chinese text categorization system based on
boosting algorithm. Proceedings of the Fourth IEEE International Conference on Computer and
Information Technology.
Chen, L., Guo, G., & Wang, K. (2011). Class-dependent projection based method for text
categorization. Pattern Recognition Letters, 32, 1493-1501.
Chen, W., Yan, J., Zhang, B., Chen, Z., & Yang, Q. (2007). Document transformation for multi-label
feature selection in text categorization. Seventh IEEE International Conference on Data Mining.
Chen, X.Y., Chen, Y., Wang, L., & Hu, Y.F. (2004). Text categorization based on frequent patterns
with term frequency. Proceedings of the Third International Conference on Machine Learning
and Cybernetics, Shanghai, 26-29 August.
Couto, T., Cristo, M., Goncalves, M.A., Calado, P., Ziviani, N., Moura, E., & Neto, B.R. (2006). A
comparative study of citations and links in document classification. Proceedings of the 6th
ACM/IEEE-CS joint conference on Digital libraries JCDL, 75-84.
Cubranic, D., & Murphy, G.C. (2004). Automatic bug triage using text categorization. Proceedings of
the Sixteenth International Conference on Software Engineering and Knowledge Engineering.
Dai, L., Hu, J., & Liu, W.C. (2008). Using modified chi square and rough set for text categorization
with many redundant features. IEEE International Symposium on Computational Intelligence
and Design.
Deng, Y.X.W.W. (2009). A New Text Categorization Method Based on SVD and Cascade Correlation
Algorithm. IEEE International Conference on Artificial Intelligence and Computational
Intelligence.
Doan, S. (2005). A fuzzy-based approach for text representation in text categorization. IEEE
International Conference on Fuzzy Systems.
Dong, T., Cheng, W., & Shang, W. (2012). The research of kNN text categorization algorithm based
on eager learning. IEEE International Conference on Industrial Control and Electronics
Engineering.
Esuli, A., Fagni, T., & Sebastiani, F. (2008). Boosting multi-label hierarchical text categorization.
Information Retrieval, 11, 287-313.
Farhoodi, M., & Yari, A. (2010). Applying machine learning algorithms for automatic Persian text
classification. Advanced Information Management and Service (IMS), 6th IEEE International
Conference, 318 323.
Ferrer, J.A., & Juan, A. (2010). Constrained domain maximum likelihood estimation for nave Bayes
text classification. Pattern Analysis and Applications, 13(2), 189196.
Fragoudis, D., Meretakis, D., & Likothanassis, S. (2005). Best terms: an efficient feature-selection
algorithm for text categorization. Knowledge and Information Systems, 8, 1633.
Frasconi, P., Soda. G., & Vullo, A. (2002). Hidden markov models for text categorization in multi-
page documents. Journal of Intelligent Information Systems, 18(2/3), 195-217.
Gao, G., & Guan, S. (2012). Text categorization based on improved Rocchio algorithm. IEEE
International Conference on Systems and Informatics ICSAI.
Guo, G., Wang, H., Bell, D., Bi, Y., & Greer, K. (2006). Using kNN model for automatic text
categorization. Soft Computing, 10(5), 423-43.
Hadni, M., Lachkar, A., & Ouatik, S.A. (2012). A new and efficient stemming technique for arabic
text categorization. IEEE International Conference on Multimedia Computing and Systems
(ICMCS).
Haifeng, L., Shousheng, L., & Zhan, S. (2010). An Improved kNN Text Categorization on Skew Sort
Condition. IEEE International Conference on Computer Application and System Modeling
ICCASM.
Harrag, F., El-Qawasmah, E., & AL-Salman, A.M.S. (2011). Stemming as a Feature Reduction
Technique for Arabic Text Categorization. 10th IEEE International Symposium on
Programming and Systems (ISPS).
Harrag, F., Al-Salman, A.M.S., & Mohammed, M.B. (2010). A comparative study of neural networks
architectures on arabic text categorization using feature extraction. IEEE International
Conference on Machine and Web Intelligence (ICMWI).
Hassan, S., Rafi, M., & Shaikh, M.S. (2011). Comparing SVM and naive bayes classifiers for text
categorization with Wikitology as knowledge enrichment. 14th IEEE International Multitopic
Conference.
Huang, J.C., Settimi, R., Zou, X., & Solc, P. (2006). The detection and classification of non-functional
requirements with application to early aspects. 14th IEEE International Conference on
Requirements Engineering, 39-48.
Ikonomakis, M., Kotsiantis, S., & Tampakas, V. (2005). Text classification using machine learning
techniques. WSEAS Transactions on Computers, 8(4), 966-974.
Jiang, S., Pang, G., Wu, M., & Kuang, L. (2012). An improved K-nearest-neighbor algorithm for text
categorization. Expert Systems with Applications, 39, 15031509.
Jo, T. (2009). Automatic Text Categorization using NTC. First IEEE International Conference on
Networked Digital Technologies NDT.
Kehagias, A., Petridis, V., Kaburlasos, V. G., & Fragkou. P. (2003). A comparison of word- and
sense-based text categorization using several classification algorithms. Journal of Intelligent
Information Systems, 21(3), 227247.
kim, J., & kim, M.H. (2004). An evaluation of passage-based text categorization. Journal of Intelligent
Information Systems, 23(1), 4765.
Ko, Y., Park, J., & Seo, J. (2004). Improving text categorization using the importance of sentences.
Information Processing and Management, 40, 6579.
Kumar, M.A., & Gopal, M. (2010). An investigation on linear SVM and its variants for text
categorization. Second IEEE International Conference on Machine Learning and Computing.
Kumar, M.A., Gopal, M. (2010). A comparison study on multiple binary-class SVM methods for
unilabel text categorization. Pattern Recognition Letters, 31, 14371444.
Lakshmi, K., & Mukherjee, S. (2005). Profile extraction from mean profile for automatic text
categorization. Proceedings of the IEEE International Conference on Computational
Intelligence for Modeling, Control and Automation, and International Conference on Intelligent
Agents, Web Technologies and Internet Commerce (CIMCA-IAWTIC05).
Lam. W., Ruiz, M., & Srinivasan, P. (1999). Automatic text categorization and its application to text
retrieval. IEEE Transactions on Knowledge and Data Engineering, 11(6),
November/December.
Lan, M., Sung, S.Y., Low, H.B., & Tan, C.L. (2005). A Comparative Study on Term weighting
Schemes for Text Categorization. Proceedings of International Joint Conference on Neural
Networks, Montreal, Canada, July 31 - August 4.
Lee, C.H., & Yang, H.C. (2009). Construction of supervised and unsupervised learning systems for
multilingual text categorization. Expert Systems with Applications, 36, 24002410.
Lee, C.H., Yang, H.C., Chen, T.C., & Ma, S.M. (2006). A comparative study on supervised and
unsupervised learning approaches for multilingual text categorization. Proceedings of the First
IEEE International Conference on Innovative Computing, Information and Control ICICIC.
Lee, K.S., & Kageura, K. (2007). Virtual relevant documents in text categorization with support vector
machines. Information Processing and Management, 43, 902913.
Lee, L.H., Wan, C.H., Rajkumar, R., & Isa, D. (2012). An enhanced Support Vector Machine
classification framework by using Euclidean distance function for text document categorization.
ApplIntell, 37, 8099.
Lee, L.H., Isa, D., Choo, W.O., & Chue, W.Y. (2012). High relevance keyword extraction facility for
bayesian text classification on different domains of varying characteristic. Expert Systems with
Applications, 39, 11471155.
Leopold, E., & Kindermann, J. (2002). Text categorization with support vector machines. how to
represent texts in input space? Machine Learning, 46, 423444.
Li, C.H., Song, W., & Park, S.C. (2009). An automatically constructed thesaurus for neural network
based document categorization. Expert Systems with Applications, 36, 1096910975.
Li, C.H., Yang, J.C., & Park, S.C. (2012). Text categorization algorithms using semantic approaches,
corpus-based thesaurus and WordNet. Expert Systems with Applications, 39, 765772.
Li, H., & Yamanishi, K. (2002). Text classification using ESC-based stochastic decision lists.
Information Processing and Management, 38, 343-362.
Li, W., Miao, D., & Wang, W. (2011). Two-level hierarchical combination method for text
classification. Expert Systems with Applications, 38, 20302039.
Li, Y., Hsu, D.F., & Chung, S.M. (2009). Combining multiple feature selection methods for text
categorization by using rank-score characteristics. 21st IEEE International Conference on Tools
with Artificial Intelligence.
Lia, Z., & Mu, J. (2010). Web text categorization for large-scale corpus. IEEE International
Conference on Computer Application and System Modeling ICCASM.
Liang, C.Y., Guo, L., Xia, Z.J., Nie, F.G., Li, X.X., Su, L., & Yang, Z.Y. (2006). Dictionary-based
text categorization of chemical web pages. Information Processing and Management, 42, 1017
1029.
Lim, B.P.C., Tsui, M.H., Charastrakul, V., & Shi, D. (2006). Web search with text categorization
using probabilistic framework of SVM. IEEE International Conference on Systems, Man, and
Cybernetics, October 8-11, Taipei, Taiwan.
Liu, L., & Liang, Q. (2011). A high-performing comprehensive learning algorithm for text
classification without pre-labeled training set. Knowledge Information System, 29, 727738.
Liu, S.Z., & Hu, H.P. (2007). Text classification using sentential frequent item sets. Journal of
Computer Science & Technology, 22(2), 334.
Liu, T., & Guo, J. (2005). Automatic text categorization based on angle distribution. Proceedings of
the Fourth International Conference on Machine Learning and Cybernetics, Guangzhou, 18-21
August.
Liu, Z., & Yang, J. (2012). An improved ambiguity measure feature selection for text categorization.
4th IEEE International Conference on Intelligent Human-Machine Systems and Cybernetics.
Lo, S. (2008). Web service quality control based on text mining using support vector machine. Expert
Systems with Applications, 34, 603610.
Lu, F., & Bai, Q. (2010). A Refined Weighted K-Nearest Neighbors Algorithm for Text
Categorization. IEEE International Conference on Intelligent Systems and Knowledge
Engineering (ISKE).
Lu, Z.Y., Lin, Y.M., Zhao, S., Chen, J.N, & Zhu, W.D. (2009). A redundancy based term weighting
approach for text categorization. World Congress on Software Engineering, IEEE.
Lucca, G.A.D., Penta, M.D., & Gradara, S. (2002). An approach to classify software maintenance
requests. International Conference on Software Maintenance (ICSM).
Luo, Q., Chen, E., & Xiong, H. (2011). A semantic term weighting scheme for text categorization.
Expert Systems with Applications, 38, 1270812716.
Malhotra, R., & Jain, A. (2011). Software fault prediction for object oriented systems: A literature
review. ACM SIGSOFT Software Engineering Notes, 36(5).
Malik, H.H., Fradkin, D., & Moerchen, F. (2011). Single pass text classification by direct feature
weighting. Knowledge Information System, 287998.
Meng, J., & Lin, H. (2010). A two-stage feature selection method for text categorization. Seventh
IEEE International Conference on Fuzzy Systems and Knowledge Discovery FSKD.
Meng, J., Lin, H., & Yu, Y. (2011). A two-stage feature selection method for text categorization.
Computers and Mathematics with Applications, 62, 27932800.
Menzies, T., & Marcus, A. (2008). Automated severity assessment of software defect reports. IEEE
International Conference on Software Maintenance (ICSM).
Miao, Y.Q., & Kamel, M. (2011). Pairwise optimized Rocchio algorithm for text categorization.
Pattern Recognition Letters, 32, 375382.
Mitra, V., Wang, C.J., & Banerjee, S. (2007). Text classification: A least square support vector
machine approach. Applied Soft Computing, 7, 908914.
namburu, S.M., Tu, H., Luo, J., & Pattipati, K.R. (2005). Experiments on supervised learning
algorithms for text categorization. IEEE Aerospace Conference, 1-8.
Nguyen, T.T., Chang, K., & Hui, S.C. (2012). Supervised term weighting centroid-based classifiers for
text categorization. Knowledge Information System.
Nuntiyagul, A., Naruedomkul, K., & Cercone, N. (2005). Recovering lack of words in text
categorization for item bank. Proceedings of the 29th Annual IEEE International Computer
Software and Applications Conference (COMPSAC05).
Nuntiyagul, A., Naruedomkul, K., Cercone, N., & Wongsawang, D. (2005). PKIP: Feature selection in
text categorization for item banks. Proceedings of the 17th IEEE International Conference on
Tools with Artificial Intelligence (ICTAI05).
Ozgur, L., & Gungor, T. (2012). Optimization of dependency and pruning usage in text classification.
Pattern Analysis and Applications, 15(1), 4558.
Pan, S., Zhang, Y., & Li, X. (2012). Dynamic classifier ensemble for positive unlabeled text stream
classification. Knowledge Information System, 33, 267287.
Pan, X., & Zhang, S. (2011). Semi-supervised fuzzy learning in text categorization. Eighth IEEE
International Conference on Fuzzy Systems and Knowledge Discovery (FSKD).
Paquet, T., Heutte, L., Koch, G., & Chatelain, C. (2012). A categorization system for handwritten
documents. IJDAR, 15, 315330.
Peng, T., Zuo, W., & He, F. (2008). SVM based adaptive learning method for text classification from
positive and unlabeled documents. Knowledge Information System, 16, 281301.
Pinheiro, R.H.W., Cavalcanti, G.D.C., Correa, R.F., & Ren, T.I. (2012). A global-ranking local feature
selection method for text categorization. Expert Systems with Applications, 39 1285112857.
Pu, W., Liu, N., Yan, S, Yan, J., Xie, K., & Chen, Z. (2007). Local word bag model for text
categorization. Seventh IEEE International Conference on Data Mining.
Qiu, L.Q., Zhao, R.Y., Zhou, G., & Yi, S.W. (2008). An extensive empirical study of feature selection
for text categorization. Seventh IEEE/ACIS International Conference on Computer and
Information Science.
Rigutini, L., Maggini, M., & Liu, B. (2005). An EM based training algorithm for cross-language text
categorization. Proceedings of the IEEE/WIC/ACM International Conference on Web
Intelligence (WI05).
Ruiz, M.E., & Srinivasan, P. (2002). Hierarchical text categorization using neural networks.
Information Retrieval, 5, 87118.
Runeson, P., Alexandersson, M., & Nyholm, O. (2007). Detection of duplicate defect reports using
natural language processing. 29th IEEE International Conference on Software Engineering
(ICSE), 499 508.
Saldarriaga, S.P., Gaudin, C.V., & Morin, E. (2010). Impact of online handwriting recognition
performance on text categorization. IJDAR, 13, 159171.
Schapire, R.E., & Singer, Y. (2000). BoosTexter: A boosting-based system for text categorization.
Machine Learning, 39, 135168.
Shanahan, J. (2001). Modeling with words: an approach to text categorization. IEEE International
Fuzzy Systems Conference.
Shi, L., Ma, X., Xi, L., Duan, Q., & Zhao, J. (2011). Rough set and ensemble learning based semi-
supervised algorithm for text classification. Expert Systems with Applications, 38, 63006306.
Sjberg, D.I.K., Hannay, J.E., Hansen, O., Kampenes, V.B., Karahasanovic, A., Liborg, N.K., &
Rekdal, A.C. (2005). A survey of controlled experiments in software engineering. IEEE
Transactions on Software Engineering, 31(9).
Song, F., Liu, S., & Yang, J. (2005). A comparative study on text representation schemes in text
Categorization. Pattern Analysis and Applications, 8, 199209.
Soucy, P., & Mineau, G.W. (2001).A simple kNN algorithm for text categorization. Proceedings of
the IEEE International Conference on Data Mining, ICDM.
Srinivas, M., Spreethi, K.P., Prasad, E.V. DR., & Kumari, S.A. (2008). MFCC and ARM algorithms
for text categorization. Proceedings of the IEEE International Conference on Computing,
Communication and Networking ICCCN.
Sun, A., Lim, E.P., & Liu, Y. (2009). On strategies for imbalanced text classification using SVM: A
comparative study. Decision Support Systems, 48, 191201.
Suzuki, M., Yamagishi, N., Tsai, Y.C., Ishidaand, T., & Goto, M. (2010). English and Taiwanese text
categorization using n-gram based on vector space model. ISITA, Taichung, Taiwan, October
17-20.
Sebastiani, F. (2002). Machine learning in automated text categorization. ACM Computing Surveys,
Vol. 34.
Tan, S., Wang, Y., & Wu, G. (2011). Adapting centroid classifier for document categorization. Expert
Systems with Applications, 38(8), 1026410273.
Tan, C.M., Wang, Y.F., & Lee, C.D. (2002). The use of bigrams to enhance text categorization.
Tao, Z.Y., Ling, G., & Chang, W.Y. (2005). An Improved TF-IDF approach for text classification.
Journal of Zhejiang University Science, 6A(1), 49-55.
Taci, S., & Gngr, T. (2008). An evaluation of existing and new feature selection metrics in text
categorization. 23rd IEEE International Symposium on Computer and Information Sciences
ISCIS.
Taci, S., & Gngr, T. (2009). LDA-based keyword selection in text categorization. 24th IEEE
International Symposium on Computer and Information Sciences ISCIS.
Torkkola, K. (2003). Discriminative features for text document classification. Pattern Analysis and
Applications, 6, 301-308.
Wan, C.H., Lee, L.H., Rajkumar , R., & Isa, D. (2012). A hybrid text classification approach with low
dependency on parameter by integrating K-nearest neighbor and support vector machine. Expert
Systems with Applications, 39, 1188011888.
Wang , Z., Sun, X., & Zhang, D. (2006). An optimal text categorization algorithm based on SVM.
Proceedings of the IEEE International Conference on Communications, Circuits and Systems.
Wang, B.K., Huang, Y.F., Yang, W.X., & Li, X. (2012). Short text classification based on strong
feature thesaurus. Journal of Zhejiang University Science C (Computers & Electronics), 13(9),
649-659.
Wang, P., Hu, J., Zeng, H.J., & Chen, Z. (2009). Using Wikipedia knowledge to improve text
classification. Knowledge Information System, 19, 265281.
Wang, T.Y., & Chiang, H.M. (2007). Fuzzy support vector machine for multi-class text categorization.
Wang, T.Y., & Chiang, H.M. (2009). One-against-one fuzzy support vector machine classifier: An
approach to text categorization. Expert Systems with Applications, 36, 1003010034.
Wang, T.Y., & Chiang, H.M. (2011). Solving multi- label text categorization problem using support
vector machine approach with membership function. Neurocomputing, 74, 36823689.
Wang, W., & Yu, B. (2009). Text categorization based on combination of modified back propagation
neural network and latent semantic analysis. Neural Computing and Applications, 18, 875881.
Wang, X., Zhang, L., Xie, T., Anvik, J., & Sun, J. (2008). An approach to detecting duplicate bug
reports using natural language and execution information. Proceedings of the 30th international
conference on Software engineering, pp. 461-470.
Wang, Y., & Wang, Z.O. (2007). A fast kNN algorithm for text categorization. Proceedings of the
Sixth IEEE International Conference on Machine Learning and Cybernetics, Hong Kong, 19-22
August.
Wermter, S. (2000). Neural network agents for learning semantic text classification. Information
Retrieval, 3, 87103.
Wilbur, W.J., & Kim, W. (2009). The ineffectiveness of within-document term frequency in text
classification. Information Retrieval, 12, 509525.
Wu, K., Lu, B.L., Utiyama, M., & Isahara, H. (2008). An empirical comparison of minmax-modular
k-NN with different voting methods to large-scale text categorization. Soft Computing, 12(7),
647-655.
Wu, Y.C., & Yang, J.C. (2012). A weighted cluster-based Chinese text categorization approach:
incorporating with word clusters. IIAI International Conference on Advanced Applied
Informatics IEEE.
Xu, J.S., & Wang, Z.O. (2004). A new method of text categorization based on pa and Kohonen
network. Proceedings of the Third International Conference on Machine Learning and
Cybernetics, Shanghai, 26-29 August.
Xu, J.S. (2007). TCBPLK: A new method of text categorization. Proceedings of the Sixth IEEE
International Conference on Machine Learning and Cybernetics, Hong Kong, 19-22 August.
Yan, B., & Qian, D. (2007). Building a simple and effective text categorization system using relative
importance in category. Third IEEE International Conference on Natural Computation.
Yang, J., Liu, Y., Zhu, X., Liu, Z., & Zhang, X. (2012).A new feature selection based on
comprehensive measurement both in inter-category and intra-category for text categorization.
Yang, Y. (1999). An evaluation of statistical approaches to text categorization. Information Retrieval,
1(1-2), 6990.
Yang, Y., & Pedersen, J. O. (1997). A comparative study on feature selection in text categorization.
Proceedings of the Fourteenth International Conference on Machine Learning ICM, 412-420.
Yen, S.J., Lee, Y.S., Ying, J.C., & Wu, Y.C. (2011). A logistic regression-based smoothing method for
Chinese text categorization. 2011. Expert Systems with Applications, 38, 1158111590.
Youn, E., & Jeong, M.K. (2009). Class dependent feature scaling method using naive Bayes classifier
for text data mining. Pattern Recognition Letters, 30, 477485.
Yun, J., Jing, L., Yu, J., & Huang, H.(2012). A multi-layer text classification framework based on
two-level representation model. Expert Systems with Applications, 39, 20352046.
Zakzouk, T.S., & Mathkour, H.I. (2012). Comparing text classifiers for sports news. Procedia
Technology, 1, 474 480.
Zelaia, A., Alegria, I., Arregi, O., & Sierra, B. (2011). A multiclass/multilabel document
categorization system: Combining multiple classifiers in a reduced dimension. Applied Soft
Computing, 11, 49814990.
Zhan, Y., & Chen, H.(2010). Reducing samples learning for text categorization. 3rd IEEE
International Conference on Information Management, Innovation Management and Industrial
Engineering.
Zhang, M.L., Pea, J.M., & Robles, V. (2009). Feature selection for multi-label naive bayes
classification. Information Sciences, 179, 32183229.
Zhang, T., & Oles, F.J. (2001). Text categorization based on regularized linear classification methods.
Information Retrieval, 4, 531.
Zhang, W., Yoshida, T., & Tang, X. (2008). Text classification based on multi-word with support
vector machine. Knowledge-Based Systems, 21, 879886.
Zhen, Z., Zeng, X., Wang, H., & Han, L. (2011). A global evaluation criterion for feature selection in
text categorization using Kullback-Leibler Divergence. IEEE International Conference of Soft
Computing and Pattern Recognition (SoCPaR).
Zheng, W., Qian, Y., & Lu, H. (2012). Text categorization based on regularization extreme learning
machine. Neural Computing and Applications, 22(3), 447-456.
Zhou, X., & Zhang, H. (2008). An algorithm of text categorization based on similar rough set and
fuzzy cognitive map. Fifth IEEE International Conference on Fuzzy Systems and Knowledge
Discovery.
Zifeng, C., Baowen, X., Weifeng, Z., & Junling, X. (2006). A new approach of feature selection for
text categorization. Wuhan University Journal of Natural Sciences, 11(5), 1335-1339.
Zifeng, C., Baowen, X., Weifeng, Z., Dawei, J., & Junling, X. (2007). CLDA: Feature selection for
text categorization based on constrained LDA. IEEE International Conference on Semantic
Computing.
Zuo, M., Zeng, G., Tu, X., & Zuo, M. (2010). Study on an Improved Naive Bayesian Classifier used in
the Chinese text categorization. Second IEEE International Conference on Modeling,
Simulation and Visualization Methods.
Appendix
AKS: Adaptive Keyword Selection, CMFS: Comprehensive Measurement Feature Selection, ARM:
Association Rule Mining, LSI: Latent Semantic Indexing, FCD: Feature Contribution Degree, CLDA:
Constrained Linear Discriminant Analysis, ALOFT: At Least One FeaTure, PKIP: Patterned Keywords In
Phrase, NBMBM: NB multivariate Bernoulli model, PSVM: Probabilistic Framework for SVM, MCMC:
Markov Chain Monte Carlo, PLS: Partial Least Square, LS-SVM: Least Square SVM, DCEPU: Dynamic
Classier Ensemble method for Positive and Unlabeled Text stream, TFKNN: Tree-Fast KNN, M3-k-NN:
Min-Max-Modular KNN, GC: Generalization Capability algorithm, CNN: Condensed Nearest Neighbor,
SNN: Selective Nearest Neighbor, RNN: Reduced Nearest Neighbor, ENN: Edited Nearest Neighbor,
BPNN: Backward Propagation NN, MRBP: Morbidity neurons Rectified BPNN, LPEBP: Learning Phase
Evaluation BPNN, CBT: Corpus Based Thesaurus, WN: WordNet thesaurus, RELM: Regularization
Extreme Learning Machine, MLP/RBF NN: Multilayer Perceptron/Radial Basis Function NN, CFC:
Class Feature Centroid.
Bibliographic information of this paper for citing:
Jindal, Rajni, Malhotra, Ruchika, & Jain, Abha (2015). "Techniques for text classification:
Literature review and current trends." Webology, 12(2), Article 139. Available at:
http://www.webology.org/2015/v12n2/a139.pdf
Copyright 2015,.

Techniques of Text Classification

Uploaded by

Copyright:

Available Formats

Techniques of Text Classification

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Techniques of Text Classification

Uploaded by

Copyright:

Available Formats

Webology, Volume 12, Number 2, December, 2015

Home Table of Contents Titles & Subject Index Authors Index

Techniques for text classification: Literature review and current trends

Received October 10, 2015; Accepted December 26, 2015

Formulation of Research Questions

Inclusion/ Exclusion criterion

Formulation of Research Questions

Figure 1. Text Classification Process

Feature Selection or Feature Transformation

Constructing a Vector Space Model

Application of a data mining algorithm

Evaluation of a text classifier

Figure 2: Confusion matrix for Category Ck

Table 1. Comparative analysis of different feature selection methods

S. Paper Technique Conclusion/ Advantage

Table 2. Comparative analysis of different document representation methods

Table 3. Comparative analysis of different data mining methods

Lo (2008) proposed a mechanism to facilitate website management, named as WebQC which

Relevant text classification journals (RQ1)

Table 4. Most important text classification journals

Rank Journal Name # Papers Prop (%) Cumulative Prop (%)

Year showing the maximum publications (RQ2)

Widely used data mining methods (RQ3)

Table 5. Most important data mining methods used

Rank Data Mining Methods # Papers

Distribution of data mining methods (RQ4)

Widely used feature selection methods (RQ5)

Table 6. Most important feature selection methods used

Rank Feature Selection Methods # Papers

Widely used datasets (RQ6)

Table 7. Most important dataset used

Rank Dataset Used # Papers

Distribution of document representation methods (RQ7)

Figure 7. Distribution of document representation methods (# papers/ percentage of total papers)

Bibliographic information of this paper for citing:

You might also like