Research On Latent Factor Model Fused With The Social And Item Information

The arrival of the information era has brought us more choices while bringing convenience to us.We can't select our required information quickly and accurately from multiple choices,which is called the problem of information overload.However,recommendation systems can help users to find satisfactory and appropriate information from a large amount of information more accurately and quickly,so it can achieve good results in solving the problem of information overload.The Latent Factor Model(LFM)is widely used as a classic recommendation algorithm.But the traditional LFM only uses the users' rating information for the items,resulting in poor recommendation quality.Because the widespread use of social,item labels,and item category information makes the recommendation system with sparse data have better accuracy,this paper integrates social and item information into LFM to describe the profile of users and items better,so that understand users' preferences and recommend suitable items for users better.A new algorithm framework proposed in this paper integrates the users' rating information of the item,the users' social relationship information and the items' information,to provide constraints for the recommendation model.The result obtained are the similar users' preference and a potential similar item,which can guarantee the quality of recommendation.The use of the information of the item can find potential similar items better to make better recommendations.Usually,the similarity between items is calculated by the items' tag vectors or genre vectors,while the sparseness of tags or genres data result in the poor similarity between items,which affect the accuracy of recommendations.The Latent Semantic Index(LSI)technology used to process items' information in this paper can describe better the similarity between items and assign different weights to items with different similarities,to find potential similar items effectively and improve the quality of recommendations.By experimenting on the Douban dataset and the Last.fm dataset,the results show that the proposed algorithm can improve the recommendation accuracy.The LSI technology only relies on the tag of the item itself,and failing to coordinate and comprehensively use the all of items' tag,so this paper also uses improved LSI technology to process items' information.the improved LSI technology globally coordinates tags and items to construct three parts: the tag frequency weight,tag local weight and the item overall weight,then combines these three parts jointly to obtain the tag-item matrix,so that utilize the tag information more effectively,and improve the recommendation accuracyeffectively.By experimenting on the Last.fm dataset,the results show that the tag information has an important influence on the recommendation quality.