Research On Citation Recommendation Based On Paper Keyphrase

With the rapid development of technology,the number of academic articles has exploded,making it more and more difficult for researchers to find the articles they need from the massive literature in the process of writing papers.The citation recommendation technology aims to intelligently find the literature list related to the researcher's query of the article from the literature database,and provide help for the researcher when querying the article.In recent years,it has attracted more and more attention from scholars.In the citation recommendation problem,the article has diversified information,and the research focus is on the content and the structure information of the citation network.Therefore,the text content and citation structure information can be combined in the citation recommendation task.In recent years,citation recommendations based on network representation learning have attracted much attention,but how to fully consider the text content and citation structure in the network is still a challenging problem.Existing work mainly focuses on the use of network topology to learn article features.However,considering the characteristics of the citation data set: the citation information of the article depends on the content of the article,and the citation information of the article can reflect the content of the article.In addition,the existing citation recommendation methods are not very interpretable,and they are not suitable for matching citations of new query articles.This article combines the citation information of the article with the text,explores the citation relationship and the internal connection of the text,and finally uses the network representation learning method for citation recommendation.The main work is divided into two parts: 1)Propose a Citation Recommendation based on Citation Relations and Text Phrases algorithm(WCN-CLE).Specifically,extract appropriate text phrases for each citation relationship by combining citation relationship information and text information(the citation relationship between A and B,related to topic X),then construct a weighted citation network,and finally use the network representation The learning algorithm learns the representation of each node and performs citation recommendation tasks.In the end,the effect of citation recommendation can be achieved,and a more reasonable explanation can be made for citation recommendation.2)Propose a New Query Article Citation Recommendation based on Keyphrase algorithm(KE-VI).Due to the use of citation structure information for network representation learning for citation recommendation,when a new query article is encountered,it is inefficient to join the network for retraining.We use all the articles in the training set as candidate articles.The algorithm can obtain the vector of the new query article,and alleviate the limitation that the key phrases of candidate articles can only be derived from their own text in the process of extracting key phrases of candidate articles.First,extract key phrases for each candidate article according to the citation structure information and text information and represent them,then select the article nodes adjacent to the candidate article node in the network structure,and select some of their key phrases as the candidate article key Supplement to the phrase.Then a weighted heterogeneous network is constructed and the vector representation of candidate articles is learned.The vector of the new query article is calculated by the similarity between the candidate article and the new query article,and then the citation recommendation is performed.This method can solve the problem of obtaining new query articles by obtaining the vector of query articles through non-training methods.The experimental results show the effectiveness of the method.