抄録
In paper recommendations, we propose a method using abstract classification to provide not only lists of papers to be read but also the reasons for the recommendations in terms of the author's viewpoints. In conventional methods, the similarity between a paper as a search query and papers that are candidates for recommendation is judged on the overall similarity, and it is difficult to provide a reason for the recommendation. In this paper, each sentence of the abstract is classified in viewpoints such as background, methods, and results, and we analyze whether the papers in the citation relationship can be classified according to each viewpoint. Then, we propose and validate methods for a paper recommendation based on the similarity of each viewpoint. The results show the effectiveness of the method based on similarity per viewpoint and the possibility of improving performance by combining it with conventional methods.