Research On Recommendation Techniques For Social Tagging Systems Based On LDA

With the rapid development of computer network, the social tagging system has become an important part of many new network applications, and it is favored by network users because of its easy to operate and easy to use. Tag recommendation is an important function of the tag system, it can provide the appropriate label to the user, which is convenient for user operation and improve the quality of tagging system, but will produce large amounts of data rich value, leading to the formation of "user resource tag" of a three-dimensional model. Existing tag recommendation method from traditional recommendation methods, mostly through the resolution of the 3D model into binary relations for processing, and this conversion process easily lead to fuzzy description of user information, semantic loss, weakening the personalized tag information problem, thereby limiting the effectiveness of personalized recommendation.In view of the shortcomings of the existing methods, based on the Cite ULike data set as the research object, put forward a kind of personalized tag recommendation model based on LDA model(LTR). The model is the use of topic models of thinking to solve problem of social tagging systems recommendation, in order to more reasonable design recommendation model, first of all to society of tag system, topic model and LDA etc. related knowledge of in-depth study. Secondly puts considering users, tags, resources and resources of the semantic information, from the semantic level of topic to mine the intrinsic link between users and resources, resources and tags, resources, construct the recommendation LTR model based on topic model of social tags.In this paper, the Cite ULike real data set of two sets of related experiments, from two points of view, respectively, to verify the effectiveness of the proposed algorithm:(1) Fixed recommended length, the comparison algorithm LTR and the current mainstream HOSVD, Folkrank, Direct Bin25, SK5, Popitem, UCTM, Majdi 's method accuracy, recall, and F1 values and verify the algorithm of recommendation effect;(2) Different density of the data set and calculation algorithm LTR in sparse data sets and relatively close data set. Recommended length from to gradually increasing process accuracy, recall, and F1 values to verify the recommended for each stage of the algorithm.The experiment results show that the LTR algorithm considering the with abundant semantic information of the text, the play covers user awareness of personalized tag to enhance the accuracy of recommendation, not only to provide users with personalized recommendation, and compared with the existing algorithms can obtain better recommendation results, extent, solves the user interest can be extracted accurately and in the case of sparse data, LTR recommended more ability.