Abstract: 【Purpose/significance】In order to help scientific research users find academic papers related to their research interests quickly and accurately, an academic paper recommendation model based on fine-grained semantic entities is constructed.【Method/process】The research topics, research objects, and theoretical and technical semantic entities identified in the early stage of the ex⁃periment are used as the content features of academic papers and core authors, and academic papers and core authors are obtained by using TF-IDF algorithm, TextRank algorithm and LDA model, respectively Use Word2vec to vectorize the feature words, and then cal⁃culate the cosine similarity between the core author and the academic paper, and recommend the top 20 cosine similarity values to the author【. Result/conclusion】Using accuracy, recall and F-values to compare and evaluate the recommendation results generated by the feature words based on the three algorithms, the results show that the feature words generated based on the TF-IDF algorithm have the best recommendation effect.The results of the recommendation are shown with examples, and it can be seen that the recommendation model proposed in this paper can more comprehensively recommend academic papers with similar research interests to scientific re⁃search users and improve the efficiency of scientific research【. Innovation/limitation】It mainly starts with the content characteristics of academic papers, and does not involve the network relationships contained in academic papers.