Paper 6
Paper 6
Paper 6
Abstract—As the amounts of online books are exponentially drastically reduced due to their busy schedules and COVID-19
increasing due to COVID-19 pandemic, finding relevant books pandemic. Instead, e-marketplaces and e-libraries became
from a vast e-book space becomes a tremendous challenge for popular hotspots. E-book reading platforms and online
online users. Personal recommendation systems have been purchasing tendencies made users discover their favorite
emerged to conduct effective search which mine related books books from many items. As a result, users tend to get swift
based on user rating and interest. Most of these existing systems and smart decisions from an unprecedented amount of choices
are user-based ratings where content-based and collaborative- using expert systems. Thus, recommendation systems came
based learning methods are used. These systems' irrationality is into the scene to customize users' searching and deliver the
their rating technique, which counts the users who have already
best-optimized results from a multiplicity of options. A
been unsubscribed from the services and no longer rate books.
This paper proposed an effective system for recommending books
personalized recommendation system was initially proposed
for online users that rated a book using the clustering method by Amazon, which contributed to raising Amazon's sales from
and then found a similarity of that book to suggest a new book. $9.9 billion to $12.83 billion in 2019 (second fiscal quarter)
The proposed system used the K-means Cosine Distance function that was 29% more than the previous year [1].
to measure distance and Cosine Similarity function to find The recommendation systems' algorithms were usually
Similarity between the book clusters. Sensitivity, Specificity, and developed based on content-based filtering [2], associative
F Score were calculated for ten different datasets. The average
rules, multi-model ensemble, and collaborative filtering.
Specificity was higher than sensitivity, which means that the
Multi-model ensemble algorithms can be used for
classifier could re-move boring books from the reader's list.
Besides, a receiver operating characteristic curve was plotted to personalized recommendation systems, but content-based
find a graphical view of the classifiers' accuracy. Most of the filtering needs a massive amount of real-world data to train the
datasets were close to the ideal diagonal classifier line and far predictive model. Apriori algorithm is used to find the
from the worst classifier line. The result concludes that association rules and degree of dependencies among rules.
recommendations, based on a particular book, are more Multiple classifiers are typical for multi-model based RS. In
accurately effective than a user-based recommendation system. that case, two different layers can be enforced. In the first
layer, a few basic classifiers are trained, and in the second
Keywords—Personalize book recommendation; layer, the basic classifiers are combined by using ensemble
recommendation system; clustering; machine learning methods like XGBoost or AdaBoost. A multi-model ensemble
algorithm is also used in spatial pattern detection. It can
I. INTRODUCTION calculate the spatial anomaly correlation with each other and
Most organizations have their recommendation system can cluster the anomaly correlations. The clustering technique
when they sell products online. But almost all the websites are works as a filter to detect spatial noise patterns [3].
not developed of the buyer interest; the organizations' force Collaborative filtering filters items based on the similar
add-on sells to buyers by recommending unnecessary and reactions. It searches a large group of people and can detect a
irrelevant products. A personalized recommendation system smaller set of users who have a similar taste for collecting
(PRS) helps individual users find exciting and useful products items. The similarity measure is a significant component of
from a massive collection of items. With the growth of the collaborative filtering. It can find the sets of users who show
internet, consumers have lots of options on products from e- the behavior to select items [4].
commerce sites. Finding the right products at the right time is Four main techniques are widely used to developed
a real challenge for consumers. A personalized recommendation systems – collaborative, content-based,
recommendation system helps users find books, news, movies, hybrid, cross-domain filtering algorithms. Firstly,
music, online courses, and research articles. collaborative filtering uses users' information and opinions to
The fourth industrial revolution emerges with a recommend products. It has narrow senses and general senses.
technological breakthrough in the fields like the internet of It can make automatic predictions based on user preferences
things (IoT), artificial intelligence (AI), quantum computing, by collaborating information from many users in a narrow
etc. The economic boom improves the living standard of sense. For example, collaborative filtering could make
people and elevates the purchasing power of individuals. predictions about a user that television shows a user like or
Nowadays, physical visits to shops and libraries have been dislike based on partial information of that user. In a general
*Corresponding Author
212 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 12, No. 1, 2021
sense, collaborative filtering involves collaborating large and used machine learning algorithms. Most of the researcher
volumes of multiple view-point, agents, and sources. It can be prefers collaborative filtering to the developed
applied in mineral exploration, weather forecasting, e- recommendation system. Collaborative filtering requires a
commerce, and web applications where a massive volume of vast amount of real-time user data that is not realistic for most
data needs to be processed to make the predictions. The recommendation systems. Besides, Table I shows that some
drawback of collaborative filtering is that it needs a researches have low accuracy, and some face overfitting due
tremendous amount of user data, which is realistic for some to small data size. In the paper, we proposed a cosine-
applications where we do not use information. distanced recommendation system that uses both user
information and preferences.
On the other hand, content-based filtering use objects
information and recommendation are made based on object Collaborative filtering is a very common technique for
similarity. Generally, content-based filtering is useful when book recommendation [18] [19] [20]. But the accuracy of this
we do not have useful information. The Similarity among the technique was 88% [21] or 89% [22], which is comparatively
products is considered while recommending. Both supervised low. However, a content-based recommendation system needs
and unsupervised machine learning algorithms are applied to an enormous amount of training data set, which is not feasible
measure the Similarity among products. The content can be for real-world scenarios [2]. When Jaccard similarity was
structured, semi-structured, and unstructured, but it must be added with collaborative filtering, it achieved the highest
synchronized into a structured format to calculate the recall. The major drawbacks of a collaborative recommender
Similarity. A hybrid recommendation system combines two or system are sparsity and cold-start issues. These issues can be
more filtering techniques to produce the output. The removed using a kernel-based fuzzy technique that scored a
performance of hybrid filtering is better comparing to 95% accuracy rate [23].
collaborative and content-based filtering. Collaborative
filtering does not consider domain dependencies, and content- The content-based filtering method [2] [24] was used to
based filtering does not consider people's preferences. A recommend items based on the Similarity among articles. The
combined effort is required from both collaborative and major drawback of this method is that it ignores current users'
content-based filtering techniques to make better predictions. ratings when suggesting new items. But user rating is relevant
The combined effort increases the common knowledge in for recommending new books or journals. As the user rating
collaborative filtering with content data and content-based information is missing in the documents, the content-based
filtering with user preferences. Cross-domain filtering filtering has low accuracy in the current book or journal
algorithms can access information that belongs to different recommendation.
domains. Cross-domain filtering algorithms make predictions Most of the systems are powered with Artificial
by exploring the source domain and increase the prediction in Intelligence that search items on popularity, correlation, and
the target domain. content of books [25]. Other popular techniques for RSs are
This paper proposed a clustering-based book listed as influence discrimination model [26], linear mix
recommendation system that uses different approaches, model [27], transfer meeting hybrid for unstructured text [28],
including collaborative, hybrid, content-based, knowledge- pseudo relevance feedback [29], fixed effect model [30],
based, and utility-based filtering. Clustering allows regrouping natural language processing with sentimental analysis [31],
all books based on the rating and user preference datasets. opinion leader mining [32], fuzzy c-mean clustering [33],
Such clustering shows remarkable prediction capability for a knowledge graph convolution network, a personal rank
personalized book recommendation system. The core target of algorithm using neural network [34], k-nearest neighbor, and
this research is to model an improved approach for frequent pattern tree [35]. Online search has an abnormal
customizing the recommendation system. effect on the recommendation system. For example, clicking
on high ranking books has no impact but clicking on low
II. BACKGROUND AND RELATED WORK ranking books has a positive impact [30]. Data sparsity is
another major problem for the traditional book
Recommendation systems (RSs) or recommendation recommendation system, which can be solved using a personal
algorithms are immensely used by personal and corporate rank algorithm using a neural network [34]. Both k-nearest
entities for searching news and information, pursuing online neighbor and frequent pattern tree are highly efficient for
shopping, engaging in social dating, executing search recommending scientific journals for academic journal readers
optimization, etc. [5] [6]. Recommendation systems escalate [35]. Moreover, several context-aware rule-based techniques
user adhesion, elevate user experience, and accelerate the use [36], and their recent pattern-based analysis [37] or
of efficiency of the system. With the rising popularity of e- classification-based techniques [38] [45] [46] or rule-based
book reading tendency, and readers increasing demands for belief prediction [39] [40] [41]can be used to build the
finding desired book, book recommendation system plays a recommendation systems. In this paper, a clustering-based
significant role [7] while choosing books. recommendation system was used to achieve the highest
Table I shows a comparison of machine learning-based accuracy.
book recommendation systems with limitations, descriptions,
213 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 12, No. 1, 2021
III. METHODOLOGY
The proposed system in Fig. 1 used a clustering technique
to develop the recommender system. Fig. 1 shows three parts
named data acquisition, preprocessing, and clustering
techniques. The datasets were collected from the Goodreads-
books repository of kaggle in this research. Though
Goodreads-books repository of kaggle contains seven datasets,
only four datasets (Books.csv, Book_tags.csv, Ratings.csv,
and Max_Rating.csv) were considered for this experiment.
The preprocessing technique was applied after merging all
datasets where we removed the lower-rated books and
developed a new dataset for analysis. Finally, the clustering
technique was applied for recommending books to those users
who stay in proximity to a specific cluster. Besides, a user can
then search for a book through a query interface, and results in
listing recommended books (Fig. 6). Fig. 1. Proposed Architecture for Book Recommendation.
214 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 12, No. 1, 2021
215 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 12, No. 1, 2021
IV. RESULTS AND DISCUSSIONS and F1 Score are 73.14%, 74.28%, and 74.18%. The
Assessment of predictive accuracy for the book sensitivity in dataset-1 is higher than other datasets, which
recommendation system is a crucial aspect of evaluation. means that the prediction probability was high for an exciting
Receiver operation characteristic (ROC) is widely used for book list. Specificity is 65% for dataset -6, which can detect
evaluating the accuracy of the classifiers [42][43]. Forecasting boring books for a reader. F-score is more useful than
is an essential part of every financial department, atmospheric accuracy. It finds harmonious relation between sensitivity and
science, and machine learning algorithms. ROC curve gives a specificity.
visual technique to summarize the accuracy of the classifiers. B. ROC Curve
It is widely used in statistical education and training.
A receiver operating characteristic curve illustrates the
A. Binary Predictor trade-off between the five different datasets' sensitivity and
For the predictions, one of the standard techniques used is specificity in Table III. It can be inferred from Fig. 4; all of
binary prediction. It contains beneficial building blocks of a our datasets have stayed close to the ideal diagonal line.
ROC curve. Every classification problem has two classes. Table IV shows Sensitivity, Specificity, F1 Score for the
Each instance (I) belongs to two sets, (P) and (N), of positive classifier. The sensitivity in dataset-1 is higher than other
and negative labels of class. A classifier instance has four datasets, which means that the prediction probability was high
possible types. If the positive instance is being classified for an exciting book list. Specificity is 65% for dataset -6,
correctly, it is considered as True Positive (TP). which can detect boring books for a reader. F-score is more
On the other hand, it is regarded as a false negative (FN) if useful than accuracy. It finds harmonious relation between
it is classified incorrectly. If the negative instance is classified sensitivity and specificity.
correctly, it is regarded as true negative (TN). Otherwise, it is C. ROC Curve
considered to be false positive (FP) if it is classified Fig. 5 presents a ROC curve that was plotted for sensitivity
incorrectly. Table II shows performance evaluation results for and specificity. Most of the datasets were closed to the
our proposed system before splitting the training dataset. The diagonal ideal classifier line. None of the datasets crossed the
test contains 1000 tuples where negative and positive tuples worst classifier line.
are 610 and 390, respectively. The proposed RS correctly
identifies 760 tuples and wrongly classifies 240 tuples. The D. User Interface
confusion matrix [44] is widely used to measure the Fig. 6 shows the user interface for the proposed system.
performance of classifiers. Table I depicts the confusion The input searching item was 'Sense and Sensibility,' a popular
matrix for this research. romantic and narrative book. As a result, the system showed
We found an FR rate (FPR), FN rate (FNR), TN rate all the similar books categorized into the romantic and
(TNR) or specificity, precision (P), recall (R), and F1 Score by narrative class.
using the following equations:
TABLE II. CONFUSION MATRIX
FPR = FP/ (TN+FP) (7)
Hypothesized Class True Class
FNR = FN/(TP+FN) (8) Positive(P) Negative (N) Total
Precision (P) =TP/ (TP+TN) (9) True (T) 490 120 610
Specificity or TNR = TN/(FP+FTN) (10) False (F) 120 270 390
F1 Score = 2 * (R * P) / (R + P) (12) TABLE III. SENSITIVITY, SPECIfiCITY, AND F1 SCORE FOR 5 DIffERENT
DATASETS
We extend this definition to include sensitivity =1-FPR
and specificity =1-FNR. Sensitivity is known as the true Datasets Sensitivity (%) Specificity (%) F1 Score (%)
positive rate, and specificity is termed as the true negative rate. 1 68.25 67.15 70.57
Table III shows Sensitivity, Specificity, F1 Score for the 2 70.25 71.15 73.55
classifier. Sensitivity calculates the proportion of desired
3 48 58 52.52
books for a user. Specificity calculates the proportion of
boring books for an individual user. F1 Score calculates the 4 73.14 74.28 74.18
harmonic mean of the desired and boring books that are 5 55 60 57.39
correctly identified. The maximum values of the F1 Score can
Average 62.928 66.116 65.642
be 1. Table III shows that the highest sensitivity, Specificity,
216 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 12, No. 1, 2021
217 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 12, No. 1, 2021
Systems Journal, vol. 11, no. 1, pp. 305-314, March 2017, doi: [21] J. Das, S. Majumder, P. Gupta, and K. Mali, "Collaborative
10.1109/JSYST.2015.2457244. Recommendations using Hierarchical Clustering based on Kd Trees and
[6] K. Lu, "Research on Intelligent Detection System for Information Quadtrees," International Journal of Uncertainty, Fuzziness and
Abnormal Defect Based on Personalized Recommendation of E-Book," Knowledge-Based Systems, vol. 27, no. 04, pp. 637-668, 2019.
in The International Conference on Cyber Security Intelligence and [22] S. Parvatikar and B. Joshi, "Online book recommendation system by
Analytics, 2019: Springer, pp. 1110-1117. using collaborative filtering and association mining," 2015 IEEE
[7] T. Y. Tang, P. Winoto and R. Z. Ye, "Analysis of a multi-domain International Conference on Computational Intelligence and Computing
recommender system," The 3rd International Conference on Data Research (ICCIC), Madurai, 2015, pp. 1-4, doi:
Mining and Intelligent Information Technology Applications, Macao, 10.1109/ICCIC.2015.7435717.
2011, pp. 280-285. [23] S. Kanetkar, A. Nayak, S. Swamy, and G. Bhatia, "Web-based
[8] S. S. Sohail, J. Siddiqui and R. Ali, "Book recommendation system personalized hybrid book recommendation system," in 2014
using opinion mining technique," 2013 International Conference on International Conference on Advances in Engineering & Technology
Advances in Computing, Communications and Informatics (ICACCI), Research (ICAETR-2014), 2014: IEEE, pp. 1-5.
Mysore, 2013, pp. 1609-1614, doi: 10.1109/ICACCI.2013.6637421. [24] A. Gholami, Y. Forghani, and M. Branch, "Improving Multi-class Co-
[9] A. S. Tewari, A. Kumar and A. G. Barman, "Book recommendation Clustering-Based Collaborative Recommendation Using Item Tags
system based on combine features of content based filtering, Improving Multi-class Co-Clustering-Based Collaborative
collaborative filtering and association rule mining," 2014 IEEE Recommendation Using Item Tags."
International Advance Computing Conference (IACC), Gurgaon, 2014, [25] E. Cho and M. Han, "AI Powered Book Recommendation System," in
pp. 500-503, doi: 10.1109/IAdCC.2014.6779375. Proceedings of the 2019 ACM Southeast Conference, 2019, pp. 230-
[10] P. Mathew, B. Kuriakose and V. Hegde, "Book Recommendation 232.
System through content based and collaborative filtering method," 2016 [26] F. Eskandanian, N. Sonboli, and B. Mobasher, "Power of the Few:
International Conference on Data Mining and Advanced Computing Analyzing the Impact of Influential Users in Collaborative
(SAPIENCE), Ernakulam, 2016, pp. 47-52, doi: Recommender Systems," in Proceedings of the 27th ACM Conference
10.1109/SAPIENCE.2016.7684166. on User Modeling, Adaptation and Personalization, 2019, pp. 225-233.
[11] S. Kanetkar, A. Nayak, S. Swamy and G. Bhatia, "Web-based [27] Y. Gao, C. Huang, M. Hu, J. Feng, and X. Yang, "Research on Book
personalized hybrid book recommendation system," 2014 International Personalized Recommendation Method Based on Collaborative Filtering
Conference on Advances in Engineering & Technology Research Algorithm," in IOP Conference Series: Earth and Environmental
(ICAETR - 2014), Unnao, 2014, pp. 1-5, doi: Science, 2019, vol. 252, no. 5: IOP Publishing, p. 052099.
10.1109/ICAETR.2014.7012952. [28] G. Hu, Y. Zhang, and Q. Yang, "Transfer meets hybrid: a synthetic
[12] Chih-Ming Chen "An intelligent mobile location-aware book approach for cross-domain collaborative filtering with text," in The
recommendation system that enhances problem-based learning in World Wide Web Conference, 2019, pp. 2822-2829.
libraries, Interactive Learning Environments", Journal Interactive [29] R. Kumar, G. Bhanodai, and R. Pamula, "Book search using social
Learning Environments, 2013, pp 469-495, DOI: information, user profiles and query expansion with Pseudo Relevance
10.1080/10494820.2011.593525. Feedback," Applied Intelligence, vol. 49, no. 6, pp. 2178-2200, 2019.
[13] S. S. Sohail, J. Siddiqui, R. Ali, "OWA based Book Recommendation [30] Q. Liu, X. Zhang, L. Zhang, and Y. Zhao, "The interaction effects of
Technique", Procedia Computer Science, 2015, pp. 126-133, doi: information cascades, word of mouth and recommendation systems on
10.1016/j.procs.2015.08.425. online reading behavior: An empirical investigation," Electronic
[14] J. Qi, S. Liu, Y. Song and X. Liu, "Research on Personalized Book Commerce Research, vol. 19, no. 3, pp. 521-547, 2019.
Recommendation Model for New Readers," 2018 3rd International [31] C. Musto, P. Lops, M. de Gemmis, and G. Semeraro, "Justifying
Conference on Information Systems Engineering (ICISE), Shanghai, recommendations through aspect-based sentiment analysis of users
China, 2018, pp. 78-81, doi: 10.1109/ICISE.2018.00022. reviews," in Proceedings of the 27th ACM Conference on User
[15] M. Kommineni, P. Alekhya, T. M. Vyshnavi, V. Aparna, K. Swetha and Modeling, Adaptation and Personalization, 2019, pp. 4-12.
V. Mounika, "Machine Learning based Efficient Recommendation [32] H. Pasricha and S. Solanki, "A New Approach for Book
System for Book Selection using User based Collaborative Filtering Recommendation Using Opinion Leader Mining," in Emerging
Algorithm," 2020 Fourth International Conference on Inventive Systems Research in Electronics, Computer Science and Technology: Springer,
and Control (ICISC), Coimbatore, India, 2020, pp. 66-71, doi: 2019, pp. 501-515.
10.1109/ICISC47916.2020.9171222.
[33] S. Vimala and K. Vivekanandan, "A Kullback–Leibler divergence-based
[16] K. Puritat and K. Intawong, "Development of an Open Source fuzzy C-means clustering for enhancing the potential of an movie
Automated Library System with Book Recommedation System for recommendation system," SN Applied Sciences, vol. 1, no. 7, p. 698,
Small Libraries," 2020 Joint International Conference on Digital Arts, 2019.
Media and Technology with ECTI Northern Section Conference on
[34] G. Wen and C. Li, "Research on Hybrid Recommendation Model Based
Electrical, Electronics, Computer and Telecommunications Engineering
on PersonRank Algorithm and TensorFlow Platform," in Journal of
(ECTI DAMT & NCON), Pattaya, Thailand, 2020, pp. 128-132, doi:
10.1109/ECTIDAMTNCON48261.2020.9090753. Physics: Conference Series, 2019, vol. 1187, no. 4: IOP Publishing, p.
042086.
[17] R. Rahutomo, A. S. Perbangsa, H. Soeparno and B. Pardamean,
[35] B. Ye, Y. Tu, and T. Liang, "A hybrid system for personalized content
"Embedding Model Design for Producing Book Recommendation,"
recommendation," Journal of Electronic Commerce Research, vol. 20,
2019 International Conference on Information Management and
no. 2, pp. 91-104, 2019.
Technology (ICIMTech), Jakarta/Bali, Indonesia, 2019, pp. 537-541,
doi: 10.1109/ICIMTech.2019.8843769. [36] Iqbal H. Sarker, "Context-aware rule learning from smartphone data:
survey, challenges and future directions," vol. 6, no. 1, p. 95, 2019.
[18] C. Sipio, D. Ruscio, and P. Nguyen, "Democratizing the Development
of Recommender Systems by means of Low-code Platforms." [37] I. H. Sarker, A. Colman, and J. J. J. o. B. D. Han, "Recencyminer:
mining recency-based personalized behavior from contextual
[19] Y. Pan, D. Wu, C. Luo, and A. Dolgui, "User activity measurement in
smartphone data," vol. 6, no. 1, p. 49, 2019.
rating-based online-to-offline (O2O) service recommendation,"
Information Sciences, vol. 479, pp. 180-196, 2019. [38] S. Hossain, D. Sarma, and R. J. J. M. L. Chakma, "Machine Learning-
Based Phishing Attack Detection," International Journal of Advanced
[20] S. B. Shirude and S. R. Kolhe, "Improved Hybrid Approach of Filtering
Computer Science and Applications(IJACSA), vol. 11, no. 9, 2020.
Using Classified Library Resources in Recommender System," in
Intelligent Computing Paradigm: Recent Trends: Springer, 2020, pp. 1- [39] S. Hossain, D. Sarma, F. Tuj-Johora, J. Bushra, S. Sen and M. Taher, "A
10. Belief Rule Based Expert System to Predict Student Performance under
Uncertainty," 2019 22nd International Conference on Computer and
218 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 12, No. 1, 2021
Information Technology (ICCIT), Dhaka, Bangladesh, 2019, pp. 1-6, [44] C. Chang, "Statistical Detection Theory Approach to Hyperspectral
doi: 10.1109/ICCIT48885.2019.9038564. Image Classification," in IEEE Transactions on Geoscience and Remote
[40] F. Ahmed, et al., "A Combined Belief Rule based Expert System to Sensing, vol. 57, no. 4, pp. 2057-2074, April 2019, doi:
Predict Coronary Artery Disease," in 2020 International Conference on 10.1109/TGRS.2018.2870980.
Inventive Computation Technologies (ICICT), 2020, pp. 252-257. [45] M. N. Alam, D. Sarma, F. F. Lima, I. Saha, R. -E. -. Ulfath and S.
[41] S. Hossain, D. Sarma, R. J. Chakma, W. Alam, M. M. Hoque, and I. H. Hossain, "Phishing Attacks Detection using Machine Learning
Sarker, "A Rule-Based Expert System to Assess Coronary Artery Approach," 2020 Third International Conference on Smart Systems and
Disease Under Uncertainty," in Computing Science, Communication Inventive Technology (ICSSIT), Tirunelveli, India, 2020, pp. 1173-
and Security, Singapore, 2020, pp. 143-159: Springer Singapore. 1179, doi: 10.1109/ICSSIT48917.2020.9214225.
[42] Y. Nieto, V. Gacía-Díaz, C. Montenegro, C. C. González, and R. G. [46] I. Saha, D. Sarma, R. J. Chakma, M. N. Alam, A. Sultana and S.
Crespo, "Usage of machine learning for strategic decision making at Hossain, "Phishing Attacks Detection using Deep Learning Approach,"
higher educational institutions," IEEE Access, vol. 7, pp. 75007-75017, 2020 Third International Conference on Smart Systems and Inventive
2019. Technology (ICSSIT), Tirunelveli, India, 2020, pp. 1180-1185, doi:
10.1109/ICSSIT48917.2020.9214132.
[43] J. Chen, X. Luo, Y. Liu, J. Wang and Y. Ma, "Selective Learning
Confusion Class for Text-Based CAPTCHA Recognition," in IEEE [47] W. Usino et al., "Document similarity detection using k-means and
Access, vol. 7, pp. 22246-22259, 2019, doi: 10.1109/ACCESS.2019. cosine distance," International Journal of Advanced Computer Science
2899044. and Applications(IJACSA), vol. 10, no. 2, pp. 165-170, 2019.
219 | P a g e
www.ijacsa.thesai.org