The Personalize Recommendation Algorithm Based On Context-aware And Neural Networks

The rapid development of the information age has led to an exponential growth of information,and human society certainly will step forward to the era of information explosion from the agricultural era or industrial era.Each person receives a huge amount of information voluntarily or passively every day,meanwhile,people have to spend precious time to find out valuable information for themselves from the sundry information,which process named information overload.In order to solve this problem,the recommendation system has come into being,and makes convenience for providing valuable information for people.The higher value things recommended,based personal preference,the better the recommendation system algorithm is.Traditional recommendation algorithms include Collaborative Filtering,Content-Based Recommendations,Hybrid Recommendations and so on.However,the traditional recommendation algorithms could not be able to meet the demand of the new environment any more with the severer problems such as sparse data,cold start and personalized recommendation.This paper attempts to optimize Singular Value Decomposition(SVD)algorithm and propose the combination recommendation algorithm based on BP neural network,the main research contents are as follows:(1)We present a new two-level SVD algorithm(TLSVD)by further dividing the user feature matrix and item feature matrix into two matrixes respectively and utilizing SVD to extract the more refined factor vectors.Concretely,TLSVD firstly divides the rating matrix into user matrix and item matrix by SVD,and then the user matrix and item matrix are further divided into two matrixes respectively through SVD.Finally,the four factor matrixes(e.g.,the user's favorite director,the type and style of the actors,and so on)are used to do prediction and recommendation.And we propose a new CSVD algorithm by employing the contextual information to improve the recommendation performance.To be specific,the recommendation results of SVD algorithm will be filtered by the contextual information(i.e.,time)so as to make the final recommendation result of CSVD related to the user's historical behavior in the same time point.We put forward a novel context-aware two-level SVD algorithm(CTLSVD)by combining the proposed CSVD and TLSVD algorithm.Specially,CTLSVD first utilize TLSVD to get the preliminary recommendation result,and then the contextual information(i.e.,time)is used to filter out the unsuitable items.In order to demonstrate the performance of our proposed algorithms,we conduct extensive experiments by comparing our algorithm with several classic algorithms(i.e.,SVD algorithm,Usersbased CF and Items-based CF).The experimental results show that the performance of CSVD,TLSVD and CTLSVD are better than those of other compared algorithms.(2)The traditional recommendation algorithms can't accurately predict the user's rating of the item due to the data sparse problem.Therefore,in this paper,the BP neural network is used to train user prediction rating based on the user-based and item-based collaborative filtering to obtain accurate rating.The experimental results show that the combination recommendation algorithm based on BP neural network,in terms of precision,recall and F1-measure,performs better than user-based collaborative filtering and the collaborative filtering algorithm based on item.