Research On Paper Recommendation Algorithm Based On Novelty And Influence

With the rapid development of network technology,the network information resources show an explosive growth trend.Academic papers are no exception.According to DBLP statistics,an average of more than 336,000 papers have been published every year since 2010.And the total number of papers published on ar Xiv in the field of artificial intelligence increased more than 20 times between 2010 and 2019.How to achieve more accurate paper search and recommendation in massive data has become a key and difficult problem in the related research field.According to incomplete statistics,most users pay high attention to the novelty of papers in the search targets of academic papers.In the traditional search and recommendation algorithms,which mainly use keyword matching queries based on context or semantics and related recommendation based on user interest model,the quality of the paper recommended is uneven and there are a lot of repetitions and redundancies.This increases the difficulty for users to consult,reduces the efficiency of users to consult,and is difficult to meet users' query needs.Therefore,this paper proposes a paper recommendation algorithm based on novelty and influence.The main research work is as follows:1.This paper proposes a novelty detection algorithm based on topic co-occurrence graph,which uses topic co-occurrence graph to simulate the relationship between thesis ideas and domain knowledge background.This algorithm extracts the change of background map as the feature set of novelty detection model,and then uses the method of combining self-coding neural network with density clustering to detect the novelty degree of a paper and give a reliability judgment on whether it is novel.The experimental results show that the novelty detection method proposed in this paper is reasonable and effective.2.This paper proposes an evaluation method of paper influence based on citation network graph,which improves the original Page Rank algorithm in three dimensions of timeliness,citation difference and academic influence potential value.The experimental results show that the high-impact papers calculated by the improved algorithm outperform the original algorithm in the two evaluation indexes of publication time and future citation.3.This paper proposes a paper recommendation algorithm based on novelty and influence,which combines novelty and influence of the paper by restarting random walk algorithm.The algorithm uses the novelty and influence of the paper as the node weight to improve the traditional citation network graph,and recalculate the probability transfer matrix,so that particles can travel to nodes with high novelty and influence with greater probability in the process of restarting random walk.The experimental results show that the proposed algorithm outperforms other comparison algorithms in terms of recall rate and NDCG.