Research And Optimization Of Context-aware Personalized Recommendation Algorithm Based On Tensor Factorization

The explosive growth in the number of available digital information and the number of Internet visitors have brought about information overload problems,which preventing users from accessing items of interest on the Internet timely.The goal of the recommendation system is to recommend items that match their interests or information they need based on user's needs and historical data.Although there is currently a lot of research in recommendation,most existing methods focus on recommending the most relevant items to users without considering any other information,such as time or peers.In fact,the user's choice of items will change with time,weather(context information)and other factors.Considering the context information in the recommendation process is helpful to generating more accurate recommendations for the user.This thesis mainly studies the influence of context information on the accuracy of recommendation.Firstly,this thesis introduces the relevant techniques and basic theories of traditional recommendation algorithm and context-aware recommendation and then based on bias tensor factorization algorithm to improve the accuracy of recommendation algorithm.In order to solve the problems in the current context-aware recommendation algorithm,two effective improved algorithms are proposed:(1)Aiming at the cold start problem of the current recommendation algorithm,this thesis associates items with the same attributes,considers item attributes in the recommendation,and proposes a bias tensor factorization context-aware recommendation algorithm integrated item attributes.For the item bias in the model,the item bias is modeled as a term related to the item attributes.And items with the same attributes share the same item bias,to study the effect of the item attributes on the user rating;Aiming at the user bias in the model,consider the effect of item attribute information on user bias,and study the user's preference for a certain type of item,in order to improve the accuracy of recommendation algorithm.(2)User ratings and item ratings fluctuate greatly over time.To address the problem of changes in ratings over time,based on the bias tensor factorization algorithm,a time-varying bias tensor factorization context-aware recommendation algorithm is proposed.Aiming at the problem that the user bias fluctuates over time,the user bias is modeled as a function that changes over time to better capture the changes in user ratings.Aiming at the problem that the item bias in the model fluctuates with time,it is believed that the item bias will not fluctuate greatly in a short time.So,this thesis divides the whole rating time into several smaller intervals,and study the changes of rating bias in each interval to predict the score more accurately.This thesis evaluates the proposed algorithms on public data sets.The experimental results show that the proposed bias tensor factorization context-aware recommendation algorithm integrated item attributes get better accuracy than other algorithms,and get better accuracy than the algorithms that consider item attributes.In addition,the time-varying bias tensor factorization context-aware recommendation algorithm proposed in this thesis gets better accuracy the latest recommendation algorithm that studies time-varying.Experiments show that the two improved context-aware recommendation algorithms proposed in this thesis both improve the accuracy of the recommendation compared with the basic algorithm.