| The rapid development of the Internet has generated a huge amount of data,which makes it difficult for people to get the desired information quickly and accurately,that is to say,the phenomenon of information overload has occurred.As a kind of information filtering technology,recommendation system can effectively solve the problem of information overload,and gradually becomes a hot research topic in related fields.And with the rapid development of social networks,the social recommendation algorithm which combines users’ social information has improved the performance of traditional recommendation algorithms to a certain extent,and has attracted the attention of researchers at home and abroad.However,the existing social recommendation Algorithms still have problems such as low accuracy of recommendation results,data sparsity,and cold start.In order to alleviate these problems,this thesis proposes a social recommendation algorithm(BSHRec)that integrates bias information and bilateral implicit relationships.The main research contents include the following parts:(1)Considering that the rating habit of users and the popularity of items are different,there will be certain bias information in the users’ rating of items,which does not reflect users’ real preference for items and will reduce the accuracy of the recommendation results.In this thesis,the average ratings of users and items are used to convert the original ratings into biased ratings during modeling,so as to learn the characteristics of users and items more accurately and comprehensively,and improve the accuracy of recommendation results.(2)Considering that there may be some noise data in users’ explicit social information,these noise data further exacerbates the problems of data sparsity and users’ cold start.In this thesis,the users’ overall similarity in social space and interaction space is used to filter out the invalid social relationship in the users’ explicit social information,and the users’ implicit social relationship is introduced into the recommendation systems through the improved Pearson similarity method.In this way,the problems of data sparsity and users’ cold start are alleviated.(3)Considering that the relationship between items is also one of the key factors affecting the recommendation results,this thesis constructs an item homogeneous relationship network from user-item interaction information,and uses it as an side information similar to users’ social relations.Adding this side information to the recommendation system is able to further alleviate the problems of data sparsity and items’ cold start.(4)The model is trained on two real public datasets Ciao and Epinions,the experimental results show that the proposed model is better than existing baseline models in the two evaluation indicators of mean absolute error(MAE)and root mean square error(RMSE).There is a performance improvement of 3%-21% in these two evaluation indicators,it indicates that the model proposed in this thesis is better than existing baseline models,and can improve the performance of the recommendation system.Figure [37] table [7] reference [71]... |
