Research And Implementation Of Recommendation Algorithm Based On Tensor Decomposition Combined With Contextual Factors

With the development of the Web 2.0,we can get a lot of information from the Internet.At the same time,people are confronted with information overload.The field of information retrieval needs a personalized recommendation system so that people can change from passive seekers to active viewers.The most commonly used recommendation algorithm is collaborative filtering(CF)algorithm.CF recommends items to user based on what items other similar users have chosen.However,it fails to provide satisfactory results.Traditional CF only considers two entities in the recommendation system:users and items.In addition of this,there are other factors influencing the recommendations,such as tag,time or weather.We call these factors contextual information.Incorporating contextual information in recommender system can provide more appropriate item recommendations.This paper proposes a tensor reduction algorithm combined with time and social relationship(TRTS).TRTS considers five-dimensional information in recommendation system:users,items,tags,time and social relationship.The algorithm reveals latent relation between users,items and tags and recommends tags for specified<user,item>object.First,this paper incorporates time factors into the construction of tensor and proposes a tensor reduction algorithm combined with time(TRT).TRT considers the dynamic change of user interests,and highlights the importance of user's recent data.In addition of this,TRTS algorithm considers social relationships to the tensor model based on TRT.TRTS researches the social relationships between users.It not only improves the accuracy of the algorithm,but also relieves the sparsity problem of data.Finally,TRT and TRTS are compared to four state-of-the-art recommendation algorithms using two real world datasets.Experimental results show that TRTS has improvements on recommended quality and recommended efficiency with two real world datasets.First,this paper analyzes the current research status of the recommendation system and model the algorithm using a tensor model.Considering the shortcomings of the existing algorithms,the time factor and social relationship are combined to the algorithm.Then,experiments of algorithm on quality,operation efficiency and parameter adjustment are carried out.The real world datasets are used to verify the advantages of the algorithm.Finally,this paper summarizes the author's work and proposes the next research direction.