Personalized Recommendation Based on Contextual Awareness and Tensor Decomposition

Introduction

With the significant improvement of Internet technology, the phenomenon of “information overload” has appeared in the Internet. Users have difficulties in quickly finding the required information among the vast amount of information. Thus, to solve the problem of “information overload” has become one of the popular research fields. Although information retrieval technology has suspended this phenomenon, it still cannot meet the service demand from the users. Using recommender system which is able to recommend service information in which users are interested according to individuals’ behavior sand preferences among a large amount of data is an effective means to solve the problem of information overload. In the research of the current recommender system, to meet the needs of the user personalized recommender system has become a popular researching direction. Traditional recommender system usually makes related calculation according to the (User-Item) two-dimensional relation, and recommends users the most relevant resources via the calculation results. Since the traditional recommendation systems don’t take the differences brought by individual preference in consideration, the technical scheme in practical applications is not comprehensive. In this case, the contextual awareness technology appears to be a new direction of research of scholars. Traditional contextual awareness recommendation algorithm combines the contextual information and establish the recommendation model based on the binary relations of “user-resources”, and then the influence of contextual information of users’ preferences will be calculated. Then it will make prediction of the users’ behaviors. JIN et al. (2012) used label information technology to solve the user interest preferences. Basing on this condition, they built a trust network, and use the random walk algorithm for top-N recommendation. But during the application process of this algorithm, it appeared the problems of the cold start and the complexity of time. Guo lei et al. (2013) discuss the asocial recommendation algorithm which had a sensitive extend of trust relationship. The degree of closeness of social networks were applied in the predicted score calculation process. It effectively improved the accuracy of recommendation. LI et al. (2015) showed in their previous work that proposed a trust network matrix decomposition algorithm based on contextual awareness, which is based on the combination of trust relations and matrix decomposition algorithm. And this method was effective in solving the problems of data sparseness degree. The SVD++ Time algorithm proposed by KOREN et al. (2012) showed an effective method to add the time contextual information to the user (s) feature vector, which can effectively solve the problem of the time variation of interest. Xiong et al. (2010) took the time contextual information as the third dimension of the third order tensor, using higher order singular value decomposition technique to modeled the dynamic change of recommend process, and produced the related recommendation list to improve the quality of recommendation. Li et al. (2011) proposed a cross domain collaborative filtering algorithm based on the differences of users' interest in different time periods. The algorithm can effectively recommend the change of user’s interest by means of tracking. Zhen et al. (2015) establishment of social tagging and temporal interest evolution model to improve the recommendation effect of micro-blog recommendation algorithm. Bin et al. (2016) They proposed collaborative filtering algorithm based on interest transfer model has a good recommendation.