Text Similarity

Abstract. Soft-cardinality spectra (SC spectra) is a new method of approximation for text strings in linear time, which divides text strings into character q-grams of different sizes. The method allows simultaneous use of weighting at term and q-gram levels. SC spectra in combination with resemblance coefficients allows the construction of a family of text similarity functions that only use the surface information of the texts and weights obtained in the same text collection. These similarity measures can be used in various tasks of natural ...

Geocoding is a method used to convert address information into geographical coordinates. It plays a vital role in displaying the relationship between geographic features and semantic information expressed in texts. The objective of this study is to reveal the quality of online geocoding from postal addresses in Turkey provided by Google Maps and Bing Maps services. The quality of geocoding services in urban areas is evaluated using two particular metrics; positional accuracy and address similarity. Positional accuracy measures the distances between point features obtained through the online geocoding and reference data. Address similarity indicates the relationship between two postal addresses based on a similarity index known as the Levenshtein distance. The same performance assessment was also made with the United States’ address data to make comparisons and discussions. The results show that services have  different geocoding capabilities in both countries because of the differen...

Location, usually defined by postal address information or geographic coordinate values, is one of the leading themes in geography. Famous global mapping services such as ArcGIS Online, Bing Maps, Google Maps, or Yandex Maps can provide users with address information of any geographic coordinates using reverse geocoding. The accuracy of retrieved addresses is quite essential for a service user. Several researchers have evaluated the accuracy of the process based on the positional errors between the retrieved and actual addresses. This article proposes a different assessment based on text similarity algorithms. In this study, the authors examine the outcomes of 15 different text similarity algorithms by comparing them with the reference data. They benefit from the binary logistic regression to evaluate the results. At the end of the case study, they conclude that the soft-term frequency/inverse document frequency algorithm is the most appropriate to measure the quality of postal addresses of all tested services. The Jaccard algorithm also produces successful results only for Google and Bing Maps services. Moreover, the study allows the reader to assess the results of reverse geocoding derived from the global map platforms that serve in the test region.

The Mongue-Elkan method is a general text string comparison method based on an internal character-based similarity measure (e.g. edit distance) combined with a token level (i.e. word level) similarity measure. We propose a generalization of this method based on the notion of the generalized arithmetic mean instead of the simple average used in the expression to calculate the Monge-Elkan method. The experiments carried out with 12 well-known name-matching data sets show that the proposed approach outperforms the original Monge-Elkan method when character-based measures are used to compare tokens.

The slowness of legal proceedings in the common law legal system is a widely known fact. Any tool which could help reduce the time taken for the resolution of a case is invaluable. Common legal systems place a great importance on precedents and retrieving the correct set of precedents is considerably time consuming. Hence, for any case whose proceedings are in progress, if there are suitable prior cases, then the court has to follow the same interpretations that were passed in the prior cases. This is to ensure that similar situations receive similar treatment, thus maintaining uniformity amongst the legal proceedings across all courts at all times. Hence, precedent cases are treated as important as any other written law (a statute) in this legal system. In this paper, we propose two new approaches to solve this information retrieval problem wherein the system accepts the current case document as the query and returns the relevant precedent cases as the result. The first approach is to calculate the document similarity using Wordnet, which is a lexical database that could be leveraged to quantify the semantic relatedness between two documents, using a semantic network. The second approach is the use of a Siamese Manhattan Long Short Term Memory network, which is a supervised model trained to understand the underlying similarity between two documents.

Penilaian Kemiripan Teks (Text Similarity) memainkan peranan yang sangat penting dalam bidang NLP (Natural
Language Processing). Dalam artikel ini, dibangun Model Vektor Kata (Word Vector Model) berbasis JST dan melatih
corpus Bahasa Cina dari Sohu News, World News, dan lain sebagainya. Artikel ini mengusulkan metode untuk
menghitung nilai kemiripan teks semantik menggunakan Word Vector Model, selanjutnya metode yang ditawarkan
akan dibandingan dengan metode tradisional seperti TF-IDF, dan hasilnya percobaan membuktikan bahwa metode
yang ditawarkan cukup efektif. Menghitung kemiripan teks merupakan metrik dalam membandingkan dua atau lebih
artikel. Pada umumnya hal ini dapat dipisahkan yaitu:
 Menghitung kemiripan semantic
 Menghitung ketidak-miripan non-semantik
Penelitian ini memiliki aplikasi yang sangat luas pada bidang Information Retrieval, Automatic Question
Answering, dan Machine Learning[1-2]. Saat ini telah banyak sekali penelitian yang matang tentang metode-metode
penghitungan kemiripan teks non-semantik, namun demikian  masih banyak hal yang dapat dilakukan untuk teks
semantik.
Pan Qianhong mengusulkan cara untuk menghitung kemiripan teks yang berdasarkan Auttribute Theory [3], dan
membangun model Properties Gravity Splitting pada suatu teks, yang menghitung korelasi antara kata kunci dengan
bantuan jarak antar titik koordinat.
Zhang Huanjiong mengusulkan metode untuk menghitung kemiripan teks berbasis Hamming Distance dan juga
menggunakan Hamming Concept [4]. Metode ini menggunakan cara baru untuk menghitung dengan kenyamanan
yang sangat baik dan akurasi yang tinggi. Metode ini merepresentasikan informasi teks dengan menggunakan kode
kata (codeword), yang memungkinkan untuk mendeskripsikan informasi teks dalam simpul/gabungan. Sementara itu
hal ini berbeda dengan menggunakan konsep spasi (vektor space) secara tradisional.
Huo Hua mengusulkan cara untuk menghitung kemiripan teks berdasarkan Compressed Sparse Vector
Multiplying, yang secara efektif mengurangi persyaratan biaya (cost of computing) dan penyimpanan (harddisk).
Hanya elemen non-zero yang disimpan dan direpresentasikan pada metode ini.
Yu Gang mengusulkan metode untuk menghitung kemiripan teks berbasis semantik leksikal [7] dengan
Algoritma Maximum Matching. Metode ini menghitung korelasi dari kedua basis vektor artikel pada semantik leksikal
dari How-Net dan mendapatkan kemiripan dari kedua artikel tersebut.

This paper presents a novel approach for building adaptive similarity functions based on cardinality using machine learning. Unlike current approaches that build feature sets using similarity scores, we have developed these feature sets with the cardinalities of the commonalities and differences between pairs of objects being compared. This approach allows the machine-learning algorithm to obtain an asymmetric similarity function suitable for directional judgments. Besides using the classic set cardinality, we used soft cardinality to allow flexibility in the comparison between words. Our approach used only the information from the surface of the text, a stop-word remover and a stemmer to address the cross-lingual textual entailment task 8 at SEMEVAL 2012. We have the third best result among the 29 systems submitted by 10 teams. Additionally, this paper presents better results compared with the best official score.

Describing, comparing and evaluating corpora are key issues in corpus-based translation and corpus linguistics for which there is still a notable lack of standards. Bearing this in mind, this paper aims at investigating the use of textual distributional similarity measures in the context of comparable corpora. More precisely, we address the issue of measuring the relatedness between documents by extracting and measuring their common content. For this purpose, we designed and applied a methodology that exploits available natural language processing technology with statistical methods. Our findings showed that using a list of common entities and a simple, yet robust and high performance set of distributional similarity measures was enough to describe and assess the degree of relatedness between the documents in a comparable corpus.

Geocoding is a method used to convert address information into geographical coordinates. It plays a vital role in displaying the relationship between geographic features and semantic information expressed in texts. The objective of this study is to reveal the quality of online geocoding from postal addresses in Turkey provided by Google Maps and Bing Maps services. The quality of geocoding services in urban areas is evaluated using two particular metrics; positional accuracy and address similarity. Positional accuracy measures the distances between point features obtained through the online geocoding and reference data. Address similarity indicates the relationship between two postal addresses based on a similarity index known as the Levenshtein distance. The same performance assessment was also made with the United States’ address data to make comparisons and discussions. The results show that services have different geocoding capabilities in both countries because of the differences in the addressing formats.

Describing, comparing and evaluating corpora are key issues in corpus-based translation and corpus linguistics for which there is still a notable lack of standards. Bearing this in mind, this paper aims at investigating the use of textual distributional similarity measures in the context of comparable corpora. More precisely, we address the issue of measuring the relatedness between documents by extracting and measuring their common content. For this purpose, we designed and applied a methodology that exploits available natural language processing technology with statistical methods. Our findings showed that using a list of common entities and a simple, yet robust and high performance set of distributional similarity measures was enough to describe and assess the degree of relatedness between the documents in a comparable corpus.

One of the main efforts of recent computational linguistics is to formalize the process of identifying and evaluating similarity between narratives, which is argued to be a key concept for all human behavior. Analyses of the data of 52 adults that participated in our empirical study offered evidence that supports the position that narrative similarity can be equated to the existence of a common summary between the narratives involved. As a further step for applying the above hypothesis, we introduced our own methods for measuring similarity of narratives through the notion of summary and compared them with some of the existing lexical-matching similarity measures. Comparisons of each computational measure with the actual similarity judgments of human participants revealed that methods that merely measure the number of shared words between two stories were unable to capture human similarity judgments, especially for stories of moderate similarity levels. However, the summary-based methods of our approach managed to reproduce human ratings for story pairs across all the range of similarity (from high similarity to low similarity).

Published in Literary & Linguistic Computing, special issue on Computational Models of Narrative (September 2014, Volume 29, Issue 3) by Oxford University Press

Full text:
http://llc.oxfordjournals.org/content/early/2014/10/10/llc.fqu046.full?keytype=ref&ijkey=SKJ9Dt7xmHI3mzp

Soft cardinality is a softened version of the classical cardinality of set theory. However, given its high cost of computing (exponential order), an approximation quadratic in the number of terms in the text has been proposed in the past. Soft-cardinality spectra (SC spectra) is a new method of approximation for text strings in linear time, which divides text strings into character $q$-grams of different sizes. The method allows simultaneous use of weighting at term and $q$-gram levels. SC spectra in combination with resemblance coefficients allows the construction of a family of text similarity functions that only use the surface information of the texts and weights obtained in the same text collection. These similarity measures can be used in various tasks of natural language processing as baseline for other methods that exploit the hidden syntactic and/or semantic structure using resources based on knowledge, inference of large corpora. The proposed method was evaluated in 30 data sets to address the tasks of information retrieval, entity matching, paraphrase and textual entailment recognition. The results raised the bar near to the best published results in the used data sets. We claim that any method that uses any resource or information external to a particular data set should outperform our method. We found that our method is an effective and challenging baseline for the evaluated tasks.

"The classical set theory provides a method for comparing ob-
jects using cardinality and intersection, in combination with well-known resemblance coecients such as Dice, Jaccard, and cosine. However, set operations are intrinsically crisp: they disregard similarities between elements. We propose a new general-purpose method for object comparison using a soft cardinality function that assesses set cardinality via an aux-
iliary anity (similarity) measure. Our experiments with 12 text matching datasets show that the soft cardinality method is superior to known approximate string comparison methods in text comparison task."

Abstract. Soft cardinality (SC) is a softened version of the classical cardinality of set theory. However, given its prohibitive cost of computing (exponential order), an approximation quadratic in the number of terms in the text has been proposed in the past. SC Spectra is a new method of approximation in linear time for text strings, which divides text strings into consecutive substrings (i.e., q-grams) of different sizes. Thus, SC in combination with resemblance coefficients allowed the construction of a family of similarity functions for text comparison. These similarity measures have been used in the past to address a problem of entity resolution (name matching) outperforming SoftTFIDF measure. SC spectra method improves the previous results using less time and getting better performance. This allows the new method to be used with relatively large documents such as those included in classic information retrieval collections. SC spectra method exceeded SoftTFIDF and cosine tf-idf baselines with an approach that requires no term weighing.

With the huge heap of data around the web, there is the need to extract information from the vast availability. This information retrieval is efficiently done by the search engines, used by millions of people regularly. Meta Search Engines finds its scope where there is a need of fused information from different search engines, as each search engine applies unique method to retrieve information. Meta Search Engines provides an interface that gives user a view of single interface but on other side there lie different search engines. From these underlying search engines, Meta Search Engine collaborate wisely the documents returned by the search engines that are chosen by database selection method. The algorithm chosen by the Meta Search Engine describes its utility i.e. more befitted the algorithm is, the more efficient is the Meta Search Engine. The motive behind this paper is to aggregate all the approaches that are somewhere discussed in literature for merging the documents returned by search engines. This provides a unique source for all the Meta search approaches used so far.

This paper presents a novel approach for building adaptive similarity functions based on cardinality using machine learning. Unlike current approaches that build feature sets using similarity scores, we have developed these feature sets with the cardinalities of the commonalities and differences between pairs of objects being compared. This approach allows the machine-learning algorithm to obtain an asymmetric similarity function suitable for directional judgments. Besides using the classic set cardinality, we used soft cardinality to ...

This paper presents a novel approach for building adaptive similarity functions based on cardinality using machine learning. Unlike current approaches that build feature sets using similarity scores, we have developed these feature sets with the cardinalities of the commonalities and differences between pairs of objects being compared. This approach allows the machine-learning algorithm to obtain an asymmetric similarity function suitable for directional judgments. Besides using the classic set cardinality, we used soft cardinality to ...

The ability to identify similarities between narratives has been argued to be central in human interactions. Previous work that sought to formalize this task has hypothesized that narrative
similarity can be equated to the existence of a common summary between the narratives involved.
We offer tangible psychological evidence in support of this hypothesis. Human participants in our empirical study were presented with triples of stories, and were asked to rate: (i) the degree of similarity between story A and story B; (ii) the appropriateness of story C as a summary of story
A; (iii) the appropriateness of story C as a summary of story B. The story triples were selected systematically to span the space of their possible interrelations. Empirical evidence gathered
from this study overwhelmingly supports the position that the higher the latter two ratings are, the higher the first rating also is. Thus, while this work does not purport to formally define
either of the two tasks involved, it does argue that one can be meaningfully reduced to the other.

Published in 2013 Workshop on Computational Models of Narrative (2013, Volume 32) by OpenAccess Series in Informatics (OASIcs)

Full text:
http://drops.dagstuhl.de/opus/volltexte/2013/4152/pdf/p129-kypridemou.pdf

The ability to identify similarities between narratives has been argued to be central in human interactions. Previous work that sought to formalize this task has hypothesized that narrative similarity can be equated to the existence of a common summary between the narratives involved. We offer tangible psychological evidence in support of this hypothesis. Human participants in our empirical study were presented with triples of stories, and were asked to rate: (i) the degree of similarity between story A and story B; (ii) the appropriateness of story C as a summary of story A; (iii) the appropriateness of story C as a summary of story B. The story triples were selected systematically to span the space of their possible interrelations. Empirical evidence gathered from this study overwhelmingly supports the position that the higher the latter two ratings are, the higher the first rating also is. Thus, while this work does not purport to formally define either of the two tasks involved, it does argue that one can be meaningfully reduced to the other.

Presented in: 2013 Workshop on Computational Models of Narrative, August 4-6, 2013 - Hamburg, Germany
Full Text: http://drops.dagstuhl.de/opus/volltexte/2013/4152/pdf/p129-kypridemou.pdf

The classical set theory provides a method for comparing objects using cardinality and intersection, in combination with well-known resemblance coefficients such as Dice, Jaccard, and cosine. However, set operations are intrinsically crisp: they do not take into account similarities between elements. We propose a new general-purpose method for comparison of objects using a soft cardinality function that show that the soft cardinality method is superior via an auxiliary affinity (similarity) measure. Our experiments with 12 ...

Building a robust MT system requires a sufficiently large parallel corpus to be avail-able as training data. In this paper, we propose to automatically extract parallel sentences fromcomparable  corpora  without  using  any  MT  system  or  even  any  parallel  corpus  at  all.  Instead,we use crosslingual information retrieval (CLIR), average word embeddings, text similarity anda  bilingual  dictionary,  thus  saving  a  significant  amount  of  time  and  effort  as  no  MT  system  isinvolved in this process. We conduct experiments on two different kinds of data: (i) formal textsfrom news domain, and (ii) user-generated content (UGC) from hotel reviews. The automaticallyextracted  sentence  pairs  are  then  added  to  the  already  available  parallel  training  data  and  theextended translation models are built from the concatenated data sets. Finally, we compare theperformance  of  our  new  extended  models  against  the  baseline  models  built  from  the  availabledata. The experimental evaluation reveals that our proposed approach is capable of improving thetranslation outputs for both the formal texts and UGC

The World Wide Web is the largest repository of public data and it is continuously expanding in size and complexity with the increasing use of internet but to retrieve the relevant documents is still a big challenge in the field of information retrieval in wide area networks. In this paper the focus is on the text and the textual similarity is measured using the different similarity functions which are used as information retrieval techniques in the field of information retrieval in wide area networks. Various similarity functions have been developed but how they are best applied in information retrieval and how similarity values should be interpreted is not answered yet. In this paper the performance modelling of information retrieval techniques using the different similarity functions or similarity measures i.e. Jaccard, Cosine, Dice and Overlap has been done in wide area networks by measuring the similarity between the keywords entered by the user in the search engine and the documents retrieved for the entered keywords. In this paper ten queries were entered for the experiment, the queries and the documents were represented using the Vector space model, Google search engine was used as search tool to retrieve the first ten documents, the Textalyser tool was used to extract the keywords from the retrieved documents and coding for the similarity calculation was done in MATLAB.