Generating a Citation Summary Based on Cited Sentences and the Implied Citation Emotions

This paper aims to generate an emotional citation summary for articles on the basis of the cited sentences and their accompanying emotions. The cited sentences to the same article were divided into three corpora according to their positive, negative or neutral emotion. Summaries were extracted from the three corpora respectively, and were then combined into a comprehensive emotional citation summary for the article. In the summary generation process, the BERT pre-training model was used to vectorize the cited sentences, Fuzzy c-means was used to cluster the vectorized citing sentences, and a LexRank + MMR combination model was applied to extract summary sentences from the clusters. The experimental results show that our algorithm outperforms all the baseline algorithms in generating citation summaries. Based on the content of the summary, the emotional citation summary was found comprehensively to feedback academic peers’ evaluation of the article, in such a way as to more comprehensively describe the function and contribution of the article in the scientific community.