Semantic Recommender System Based on Semantic Similarity Using FastText and Word Mover’s Distance

Abstract

Lecturers and students use research papers as references for their projects, such as research, final projects or theses. Obtaining suitable papers is critical for the success of their projects. Nonetheless, finding the correct papers, especially in determining the correct words to paper searching, is challenging. This paper proposes a method that combines FastText and Word Mover's Distance for a recommender system to solve the challenge. FastText and Word Mover's Distance are utilized to gather word embedding and reach semantic similarity, respectively. Data on papers from the IEEE Xplore Digital Library, ACM Digital Library, Science Direct, SpringerLink, and Wiley Digital Library are collected as case studies in this paper. Experiments for verifying the proposed method referring to seven scenarios in which additional three types of queries. The first query applies the original word from the case studies, whereas the second query operates the original word with one word or two words chosen by a user. The third query uses the original word-combining with chosen words based on the proposed method. After implementing these scenarios, most of the results indicate that results produced by the proposed method gain higher accuracy and F-1 score than user query and original query. The proposed method increases a mere 2.3 percent of accuracy and 1.7 percent of F-1.