Academia.eduAcademia.edu

Figure 6 – uploaded by Talha Bin Sarwar

See full PDFdownloadDownload figure
Table 7 Obtained top five similar news articles for a particular article employing KP-Miner with Jaccard similarity measure

Table 7 Obtained top five similar news articles for a particular article employing KP-Miner with Jaccard similarity measure

Related Figures (11)

Figure 1 Word2Vec training model skip-gram. This model takes a word as input and tries to predict the similar words based on its context. Full-size Mal DOT: 10.7717/peerj-cs.1024/fig-1 similarity measure is used to calculate the similarity after the word vector values generated
Figure 1 Word2Vec training model skip-gram. This model takes a word as input and tries to predict the similar words based on its context. Full-size Mal DOT: 10.7717/peerj-cs.1024/fig-1 similarity measure is used to calculate the similarity after the word vector values generated
Figure 2 Functional details of the proposed methodology for finding similar articles.
Figure 2 Functional details of the proposed methodology for finding similar articles.
Table 1 Example of extracted keyphrases using different algorithms for a particular paper. lable 2 Fleiss’ values for the extracted keyphrases scored by the annotators.
Table 1 Example of extracted keyphrases using different algorithms for a particular paper. lable 2 Fleiss’ values for the extracted keyphrases scored by the annotators.
Figure 3 An example of word clouds generation with extracted keyphrases by employing different algorithms, namely KEA (A), MR (B), TR (C), KP-Miner (D), YAKE (E), and TeKET (F) for the comparative analysis. Full-size MY DOI: 10.7717/peerj-cs.1024/fig-3
Figure 3 An example of word clouds generation with extracted keyphrases by employing different algorithms, namely KEA (A), MR (B), TR (C), KP-Miner (D), YAKE (E), and TeKET (F) for the comparative analysis. Full-size MY DOI: 10.7717/peerj-cs.1024/fig-3
Figure 4 Comparison of the employed keyphrase extraction algorithms along with different similarity calculation techniques. Full-size MY DOI: 10.7717/peerj-cs.1024/fi After the keyphrases are extracted, they are used for the similarity calculation. The imilarity calculation part is essential in this study because it finds similar messages base mn the similarity score. The similarity is calculated based on the extracted keyphrases b ising the best performing keyphrase extraction algorithms except for TeKET since eKET has already been disregarded due to its poor performance. Similarity is compute oth lexically and semantically. For the lexical similarity calculation, the cosine and accard measures are used. For semantic similarity computation, on the other hand, the concept of word embedding is used by employing Word2Vec with cosine similarity. In ‘ig. 4 you can see a comparative analysis of the different techniques used for similarity computation along with the algorithms used for keyword extraction. Figure 4 represent he average similarity scores of the top five most similar news articles for a given news ticle for which the keyphrase extraction algorithms with different similarity measures vere used. The top five articles are acquired for the news article titled “It’s not just Delta yther coronavirus variants worry scientists’.
Figure 4 Comparison of the employed keyphrase extraction algorithms along with different similarity calculation techniques. Full-size MY DOI: 10.7717/peerj-cs.1024/fi After the keyphrases are extracted, they are used for the similarity calculation. The imilarity calculation part is essential in this study because it finds similar messages base mn the similarity score. The similarity is calculated based on the extracted keyphrases b ising the best performing keyphrase extraction algorithms except for TeKET since eKET has already been disregarded due to its poor performance. Similarity is compute oth lexically and semantically. For the lexical similarity calculation, the cosine and accard measures are used. For semantic similarity computation, on the other hand, the concept of word embedding is used by employing Word2Vec with cosine similarity. In ‘ig. 4 you can see a comparative analysis of the different techniques used for similarity computation along with the algorithms used for keyword extraction. Figure 4 represent he average similarity scores of the top five most similar news articles for a given news ticle for which the keyphrase extraction algorithms with different similarity measures vere used. The top five articles are acquired for the news article titled “It’s not just Delta yther coronavirus variants worry scientists’.
Table 3 Different categories with relevancy scores to rank similar news articles. Table 4 Performance comparison of the employed different algorithms along with similarity techniques for finding similar articles.
Table 3 Different categories with relevancy scores to rank similar news articles. Table 4 Performance comparison of the employed different algorithms along with similarity techniques for finding similar articles.
top five articles identified by different algorithms are manually ranked by the experts by giving them relevancy scores. The ranking made by the experts based on the relevancy scores is considered as the benchmark. The same group of postgraduate students has also participated in the ranking process to give relevancy scores to the top five similar articles identified for a given target article by applying various keyphrase extraction algorithms and similarity calculation techniques. Table 3 shows the relevancy scores, which are divided into four categories.
top five articles identified by different algorithms are manually ranked by the experts by giving them relevancy scores. The ranking made by the experts based on the relevancy scores is considered as the benchmark. The same group of postgraduate students has also participated in the ranking process to give relevancy scores to the top five similar articles identified for a given target article by applying various keyphrase extraction algorithms and similarity calculation techniques. Table 3 shows the relevancy scores, which are divided into four categories.
Table 5 Obtained top five similar news articles for a particular article employing KP-Miner with Cosine-Word2Vec similarity measure.
Table 5 Obtained top five similar news articles for a particular article employing KP-Miner with Cosine-Word2Vec similarity measure.
ble 6 Obtained top five similar news articles for a particular article employing KP-Miner with Cosine similarity measure.
ble 6 Obtained top five similar news articles for a particular article employing KP-Miner with Cosine similarity measure.
Figure 5 Performance of the proposed approach (employed keyphrase extraction algorithms along with different similarity calculation techniques) for finding similar news articles. Full-size My DOI: 10.7717/peerj-cs.1024/fig-5
Figure 5 Performance of the proposed approach (employed keyphrase extraction algorithms along with different similarity calculation techniques) for finding similar news articles. Full-size My DOI: 10.7717/peerj-cs.1024/fig-5
Figure 6 Comparison of different similarity measures for top five news articles obtained by the KP-Miner algorithm. Full-size My DOI: 10.7717/peerj-cs.1024
Figure 6 Comparison of different similarity measures for top five news articles obtained by the KP-Miner algorithm. Full-size My DOI: 10.7717/peerj-cs.1024
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved