Research On Recommendation Algorithms Based On Context And Tag Relevance

Context-based and Tag-related Recommendations mean that the next item of users' interest is predicted by using the information of context and tags.With the maturity of information technology,many kinds of data information,such as attribute tag of users,time-context,rating,evaluation,feature tag of items,are more and more easily accessible.Thus,how to effectively integrate the context and tag information acquired to exploit and predict user preferences is particularly important,and has become a hot research direction of recommendation systems.At present,the research on context-based recommendation algorithm and tagbased recommendation algorithm has made some progress,but there are still some problems: Firstly,when pursuing the recommendation accuracy rate,the diversity of user interests is often ignored,which makes the recommendation results homogeneous;Secondly,although the influence of temporal information has been noticed,the factor of time interval has not been fully considered,that affects the accuracy of recommendation results.Finally,with its sparsity degree of more than 90 percent,the rating data is very sparse.Recommendations algorithm that rely on very sparse rating data reduce their effectiveness.In view of the above problems,the time-context and tag-related which affect the recommendation effect are combined in this paper,and a new recommendation algorithm is proposed to improve the recommendation effect.(1)Focusing on tag information and temporal information to alleviate the problem of "data sparsity" and improve the diversity of recommendation,the following improvements are made in this paper: Firstly,user preferences are classified by integrating user attribute tag information and rating information,and item relevance is calculated more accurately by combining feature tag information of items,the diversity of recommendation results is also improved by mixing user-preference sets and itemrelated sets as the prediction benchmark.Then,tag matrices of user and item are reconstructed by fusing the popularity of tags and time parameters,thereby the information of the two tag matrices is fused to ensure the trust of similarity values,and the recommendation results are obtained.Through comparative experiments,it is found that the new algorithm improves the diversity of recommendation while guaranteeing the accuracy of recommendation.(2)Making full use of tag correlation and considering the influence of time interval,a new algorithm with fusing session segment(time interval)and tag information is proposed.In order to facilitate model of time interval information,the validity of Long Short Term Memory networks for modeling sequence data information has been considered.Firstly,session segments are introduced into the Long Short Term Memory networks model to study the effect of time interval on tags.Then,each output layer is combined with LDA topic model to weigh the tags with high importance.Finally,the prediction value is obtained by fusing the rating information.Experiments show that the new algorithm alleviates the problem of data sparsity and improves the accuracy of recommendation results.