Research On Hybrid Recommendations Based On Social Tagging

Personalized recommendation can predict the items that users are interested in according to their historical behavior preference information,and actively recommend them to users,effectively solve the problem of information overload.The content-based personalized recommendation relies on item characteristics,for the platform with various types of items to be recommended,the traditional method can only recommended according to the characteristics of the item type,resulting in a single type recommendation.However,social Tagging can not only describe the content of items but also reflect the user's preferences,which has become an important characteristic of connecting users and items.The use of social tagging as the recommendation basis can ensure diversified types of recommendation results and truly achieve content-based recommendation.Therefore,this paper studies the recommendation method based on social tags,the main contents are as follows:1.the tag-item diffusion method TID is proposed,which is based on the content recommendation of target users.The method transforms the recommendation problem into a graph search problem,and constructs the tag-item two-part graph,which lies in extracting the item feature from the new aspect as well as highlights the implicit relationships among the items.However,there exists some noise in the social tagging data,and it will lose a proportion of information if it is pre-treated.In addition,popular tags can lead the overfitting problem in user preference model.Therefore,this paper utilises bayesian classifier to weight tags and items,and the weighted tag-item diffusion method WTID is put forward,so as to improve the recommendation accuracy.2.The WTID-collaborative filtering series recommendation model is proposed,and the WTID algorithm is combined with the user-based collaborative filtering algorithm to conduct pipelined mixed design,effectively solves the defects of single algorithm.The WTID algorithm solves the problem of cold startup in personalized recommendation and improves the recommendation efficiency through coarse recommendation;collaborative filtering algorithm can take advantage of the group cooperation power and solves the problem of single item categories in coarse recommendation.The implicit user preference based tags is introduced in the nearest neighbor solution.In this way,the user preference model is improved and the dimension of nearest neighbor solution is increased,in order to make full use of the two kinds of data sources referring to user rating and the user's tagging.The two parts of the model complement each other in order to achieve high-precision personalized recommendation.3.In this paper,through the multi-group comparison experiments of Movielens dataset,it can be verified that the recommendation results of WTID algorithm and WTID-collaborative filtering series recommendation model have improved in the aspects of the value of F1 and diversity.Finally,WTID-collaborative filtering series recommendation model is applied to argumentation system and its application effect was analyzed.The results display that the model can effectively improve the research efficiency by recommending personalized information for experts.