Abstract: 【Purpose/significance】When a paper is cited many times, the cited topics and key points are different due to different cita?
tion motives. Identifying these cited topics and analyzing their changes will help analyze the citation motives and improve the paper
recommendation effect.【Method/process】Firstly, we extract the context information of the cited reference and define various scope of cited content based on text length. Then we identify cited topics and compare their similarities and differences through various text
clustering methods. Finally, we analyze the evolution path and process of cited topics through time series comparison.【Result/conclu?
sion】After analyzing the representative highly-cited papers in the field of artificial intelligence, it is found that the preceding and fol?
lowing sentences are important supplements to the current cited sentences, and the cited sentences and their combination of preceding and following sentences can better represent the cited topics. The cited topics based on the citation content are diverse, revealing the differences between the expansion of the original content and the motivation of citation. Evolutionary analysis of cited topics can effec? tively reveal the direction, theme, methods and techniques of applying or improving the original content.【Innovation/limitation】This paper forms a framework for the identification and evolution analysis of cited topics of papers based on citation content clustering to prove the differentiation of cited topics and their supplement to the original text, and at the same time, to reveal the characteristics and practical significance of the temporal changes of cited topics. Future research need expand the research samples and make the re? search results more universal.