Research On The Application Of Subject Term Recommendation Based On Data Bias Elimination

With the wide application of recommender system in the Internet,more and more scholars are trying to explore ways to effectively integrate with one-class collaborative filtering for recommender systems in more fields.In recent years,the recommendation service in the fields of literature service and academic exchange has gradually become a research direction,such as literature resource recommendation,scholar recommendation,subject term recommendation,etc.The study of data bias in literature recommendation services from the perspective of scholar behavior has also become a research point.The problem of data bias will have many adverse effects on the subject term recommendation service for scholars and literature.For example,selection bias will cause scholars to select and use only a small part of the massive subject terms in the literature,resulting in a very obvious data sparsity problem,and the literature lacks a sufficient number of subject terms to express the scholar’s preference.Conformity bias can cause scholars to be influenced by recent major research topics when writing academic papers,and make scholars use frequency of some specific subject terms in the paper different from their own preferences and the actual importance of the subject terms to the literature.This paper mainly focuses on these two common data bias problems in the field of subject term recommendation,as follows:On the one hand,in order to deal with the problem of selection bias,this paper constructed a matrix decomposition model based on literature richness and subject term popularity in order to measure the correlation probability of literature and subject terms that do not appear in the current literature,and divided these subject terms into implicit related subject terms and implicit unrelated subject terms of literature according to the correlation probability.For these two kinds of subject terms,two different weight prediction methods of subject terms were proposed,namely,Auto Rec Filling with Preference Coefficient and Zero Filling.Experiments showed that these two methods can respectively improve the recommendation effect of predicting the weight of subject terms and identifying high weight subject terms.On the other hand,in order to deal with the problem of conformity bias,this paper proposed a recommendation model that combines the elimination of conformity bias and the personalized preference of literature.Firstly,we constructed an unbiased model,which eliminates the negative impact of conformity bias at the data level by means of weight penalty.Secondly,we done the elimination of conformity bias operation in explicit weights,and combined with the matrix decomposition algorithm based on neural network,then we mined the personalized preference characteristics of the literature to the subject terms,and finally integrated the personalized preference characteristics of the literatures into the previous debiased model.Experiments showed that the proposed method can effectively eliminate the negative impact of conformity bias and improve the accuracy of subject term recommendation for scholars and literature.The experimental results showed that the research methods of this paper can effectively deal with the problems of selection bias and conformity bias in the subject term recommendation service,and significantly improved the recommendation effect in different subject term recommendation scenarios.