Application of k-means clustering algorithm to improve effectiveness of the results recommended by journal recommender system

• Published in: Scientometrics Vol. 127; no. 6; pp. 3237 - 3252
• Main Authors: Vara, Narjes, Mirzabeigi, Mahdieh, Sotudeh, Hajar, Fakhrahmad, Seyed Mostafa
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.06.2022; Springer Nature B.V
• Subjects: Algorithms; Cluster analysis; Clustering; Computer Science; Feasibility studies; Information Storage and Retrieval; Library Science; Recommender systems; System effectiveness; Vector quantization; Websites; Journal recommender systems; k-means clustering algorithm; RICeST journal finder system; Effectiveness; Scalability
• ISSN: 0138-9130; 1588-2861
• DOI: 10.1007/s11192-022-04397-4

This study investigates to evaluate feasibility of k-means clustering algorithm in order to improve effectiveness of the results recommended by RICEST Journal Finder System. More than 15,000 papers published in filed of engineering journals during 2013–2017 were collected from their websites. Their titles, abstracts and keywords were extracted, normalized and processed in order to form the test body. According to the number of papers collected, using Cochran's formula, 400 papers completely relevant to the subject of each journal were randomly and proportionally selected and entered the system as queries in order to receive the journals recommended by the system before and after k-means clustering algorithm and the results were recorded. Finally, effectiveness of the system results was determined at each stage by leave-one-out cross validation method based on precision at K top ranked results. Also, opinions of subject reviewers on relevance of the target journal were investigated through a questionnaire. Results showed that before data clustering, only 40% of target journal was recommended at the first 3 ranks. But after k-means clustering algorithm, in more than 80% of searches, the target journal was retrieved at the first 3 ranks. Also, effectiveness of the recommendations, according to 210 subject reviewers, after k-means clustering algorithm, showed that more than 80% of the recommended journals are completely relevant to the given paper. According to the study results, data clustering can significantly increase effectiveness of the results recommended by journal recommender systems.