Research On Recommendation Based On Context-aware And User Behavior

With the development of mobile internet,the information overload problems faced by information producers and information consumers are getting more and more serious.The recommendation system alleviates this problem,so that people can find the useful information from the massive information more accurately and timely,therefore,it has been widely used in many fields such as social platforms and e-commerce.With the number of users increased,the complexity of social relationships,and the diversification of information,the traditional recommendation algorithms cannot effectively satisfy the user's needs.Therefore,it is of great significance to further study the recommendation algorithms to improve the recommendation quality of the recommendation system.In summary,through studying the current lack of recommended algorithms,this thesis proposes two methods,one is a friend recommendation method which based on social networks and user interests,the other is a rating prediction method which based on social context and tensor decomposition,and the main work content is as follows:1.This thesis designs a friend recommendation method based on social networks and user interests.Firstly,according to social relations and historical rating behaviors,the proportion of users' common friends and the proportion of users' common behaviors are calculated separately.Based on this,a similarity matrix of user is constructed.Secondly,using tag data and the characteristics of TF-IDF in text information mining to build user affinity model.Finally,by giving different weights to the above two similarity degrees,we fuse the social and interest similarity degree to get user similarity degree,finally generate similar friend recommendation lists for the target users.2.This thesis designs a rating prediction method based on social context and tensor decomposition.Firstly,taking into account the propagation effect of user interest in social networks,using the user's explicit attributes and user's implicit attributes respectively to establish a rating prediction model based on cloud model,then using Gauss transform fuses two-part rating prediction results and generate predicted ratings.Secondly,due to the characteristics of tensor factorization model in data dimension conversion and data compression,applying tensor factorization model for adding contextual information to iterative update rating prediction results.Lastly,the final predicted ratings are generated by the result of compromising context-based rating prediction and user behavior-based rating prediction.Finally,this thesis uses douban data to validate the proposed friend recommendation method and the rating prediction method.