Research On The Prediction And Recommendation Of High-influential-articles

In the era of information explosion,the number of academic literature presents a rapid growth trend,which brings an increasingly prominent problem of information overload to the academic community.At the same time,it also brings challenges for researchers to obtain valid academic papers.For researchers,if they can find the high impact papers that meet their needs in the vast scientific research literature as soon as possible,it will greatly improve the efficiency of literature search,and speed up their cognition of the topic with the help of reading the high impact papers.Academic paper recommendation is an effective way to solve the problem of academic literature overload.It aims to automatically select the academic literature that researchers are interested in from the huge literature database.However,the existing academic paper recommendation systems usually implement recommendation from the perspective of content similarity and user similarity.For this reason,this recommendation method ignores the researchers' citation preference when quoting references,which leads to the effect of recommendation not good.The research of this paper is divided into two progressive processes.Firstly,aiming at the problem of which kind of characteristics of articles are more likely to be cited,the paper introduces the topic characteristics of the article,and verified that the topic characteristics of articles are the key factors affecting the citation of articles.Secondly,with the help of the key factors including the topic characteristics that affect the citation characteristics of articles,the paper recommendation model is constructed to implement literature recommendation for researchers.(1)This paper quantifies the topic characteristics of academic papers and explores the influence of topic characteristics on citation characteristics.In this paper,LDA topic model is introduced to quantify the topic characteristics of academic papers by investigating the probability of the content of academic papers belonging to the current hot topics.Combined with the topic characteristics,early citation characteristics,bibliometric characteristics of the article and the authors' characteristics,the feature space of citation behavior prediction is constructed.Three feature selection methods,Fisher score,Relief-F and SPEC,are used to sort and filter the features.SVM,KNN and Bagging are used to verify the effect of the core features in identifying future high impact papers.The experimental results show that the topic characteristics of academic papers do have an important impact on the citation behavior of papers,and its addition helps to achieve the prediction of high impact literature.In fact.it actually indicates a content tendency of researchers when they cite papers.In other words,the content of papers is the key factor to determine whether they are cited by researchers.(2)Combined with the key characteristics of the paper,such as the topic characteristics,which affect the citation behavior of the paper,a multi-dimensional information driven Top-N paper recommendation model is constructed.LDA topic model is used to extract the topic features of target papers and candidate papers,and candidate papers are preliminarily screened according to the topic features.Candidate papers with low topic similarity with target papers are excluded,and a preliminary list of candidate recommended papers is generated.The paper comprehensively measures the researchers' interest in the papers in the candidate list from the five dimensions of article preference,reputation preference,content preference,citation behavior preference and citation network preference,sorts the candidate papers according to the order of interest value from large to small,and recommends the Top-N papers to the researchers.Experimental results show that,compared with the baseline recommendation methods based on PageRank and text similarity,the multi-dimensional information driven Top-N paper recommendation model achieves better recommendation performance.The research of this paper points out that the topic characteristics of the article is an important factor affecting the future citation characteristics of the article,which determines the researchers' interest in the article to a certain extent.At the same time,it is helpful to improve the effect of literature recommendation by considering the topic characteristics and citation preference.This provides an effective way to develop a better recommendation system and provide more targeted paper recommendation service for researchers.