An Item Recommendation Method Based On Tag Quality Mining

In the recommended system,users and items are two core entities.In the field of social recommendation,tag is another important entity other than the user and the item.Tags in social recommender systems enable users to efficiently describe,classify,andnavigate vast collections of items.From this point of view,the tag is the bridge between the user and the item,high quality tags can provide useful and interesting information in item recommendation,by taking advantage of the user-tag-item relations.Much effort has been devoted to exploit tag quality in the ternary relations.However,few researchers are concerned about the quality of the tag mining.Moreover,there have been studied the majority of the ternary relationship,the distinction between the user-tag and tag-item difference between the two binary relations.In view of the shortcomings of the existing research,this paper presents a novel method of tag quality mining for auxiliary items.This method is accomplished by digging the high quality independently in two spaces.The specific work of this paper is:1?In the tag-item subspace to dig the tag leader.Among them,the tag leader is a high-quality tag of the tag-item subspace.Then,for each label to determine the tag leader,these tags become the tag leader subordinate label.Finally,the tag leader probabilistic model is established to raise the quality of the tag by tag leader and subordinate tags to raise the low quality tags.2?Explore the tag ruler in the user-tag subspace.First,in the subspace to eliminate redundant rules,non-redundant rules in the front of the label is standardized.Then,establish the tag ruler probabilistic model.Finally,by standardizing the tag to tap the user's potential interest,get the user more interest in the label.3?The tag-item subspace and the user-tag subspace are combined together by mining the tag,and establish the composite probabilistic model.Find the user the binary relationship between the items,through the ComProb to achieve the final items recommended.Experiments on the Last.FM and CiteULike datasets to start,the results show that the subspace of the label quality mining to improve the recommended performance of items have a positive impact.