Research On Technology For Context-Aware Recommendation

Personalized recommendation system aims to solve the problem of information overload,has been widely applied in various fields of the Internet.Traditional recommendation system only focus the relation between users and items to provide users with recommendations,and ignore the influence of context information to the user decision.As context-awareness technology and the rapid development of the intelligent mobile terminal technology,the research of context-aware recommender system is becoming intensive.And the research on information retrieval,mobile Internet,Internet of Things,e-commerce,smart home/office/traffic,and many other industries has extensive application prospect.Current research in the field,there still are some problems with context information mining and testing,user modeling and behavior analysis,user preferences extraction and recommendation algorithm.In order to further improve the recommendation accuracy and efficiency of recommendation,we took a further study about context modeling and recommender algorithm in this paper,and we obtained the following results:(1)To get more specific rating data,a context-based complex splitting approach based on binary particle swarm optimization is proposed to further improve the accuracy of recommendations.In our method the user or item are split into two users or items rated in different contextual conditions.The method uses discrete binary particle swarm optimization algorithm to optimize the best context condition combination for splitting.And then,one item or user are split into two different items or users according to these optimal context conditions in combination,one is rated in context that meets all context condition of best context combination,and the other one is rated in context that not meets.So more specific rating data are obtained,that will results more accurate recommendation when entered to recommendation algorithm.We evaluated our algorithm on a real world dataset and experimental results demonstrate its validity and reliability.(2)In view of the existing related research seeing all contexts as equally important,a new contextual modeling algorithm based on bayes method and clustering is proposed in this paper.This paper first cluster items using feature clustering method,and then use the bayesian formula to calculate the probability of a user liking items in a particular category with the single context conditions,so the joint probability of the user liking this kind of items with multiple context conditions is obtained.Finally,because the similarity between users those like items in same category as well should be higher,the joint probability above is incorporated into traditional collaborative filtering algorithm to improve the user similarity computing,which is beneficial to the improvement of rating prediction accuracy.Results of quantities of comparison experiments with a real world dataset demonstrate that,compared to traditional collaborative filtering method,the proposed algorithm can take full advantage of context information to improve the accuracy of recommendation effectively.