Research On Citation Recommendation Based On Heterogeneous Network Representation Learning

With the vigorous development of scientific research,the number of scientific research documents has grown rapidly.Researchers need to spend a lot of time and effort to retrieve literature related to their research field when writing an article.Citation recommendation can recommend references for scientific researchers in the process of writing articles and improve the researcher's literature effectively and accurately.Although the traditional citation recommendation method can meet the recommendation requirements to a certain extent,there are problems such as cold start and low recommendation efficiency.The academic network is a typical heterogeneous information network,which contains rich semantic features and structural features.It is of great theoretical and practical significance to study and analyze the academic network.Network representation learning is widely used in the field of information network analysis,and can represent network information with low-dimensional vectors.This paper studies the citation recommendation by constructing heterogeneous academic networks based on network representation learning methods.This paper first describes the background and importance of citation recommendation research,and reviews the current research situation at home and abroad,and analyzes the advantages and disadvantages of existing citation recommendation methods at the same time.This paper introduces the theory and mainstream models of network representation learning,and the calculation of information network entity similarity.In the context of citation recommendation,the keywords,authors,and articles of the candidate document collection constitute an academic heterogeneous network.Using the network representation learning method can effectively mine its semantic and structural features,and obtain a vector representation of each node in the network.Next,this paper constructs a citation recommendation framework based on network representation learning,and introduces the principles and practical steps of each process in detail.This paper proposes a method for constructing the target document vector from the keyword and author node vectors based on the network representation learning method to obtain the articles,keywords and author vectors in the candidate documents.This paper calculates and sorts the cosine similarity between the two vectors to return the references.Finally,on the citation data sets of AAN,MED and DBLP,this paper analyzes the recommendation effect of the citation recommendation method based on network representation through experiments,and proves the rationality and effectiveness of the method.