Weibo Label Recommendation Based On Local Label Propagation And Co-occurrence

With the rapid development of social networking services such as Sina Weibo and Twitter,more and more users express their personal knowledge,feelings,opinions and comments on these social platforms,leading to explosive growth of user scale and innovation and change of social service mechanism.Millions of users publish tens of millions of Weibo every day,corresponding to tweets on Twitter,generating rich and varied user content on various social media,how to quickly and accurately find users' interested content has become a major direction for researchers.So we have designed a comprehensive recommendation algorithm based on the social relationship between users,the co-occurrence relationship between the labels and the semantic relationship,which is unique in its comprehensiveness.At first,Our experiments have shown that the interaction between users and the relationship between friends can indeed explain the similarity of interest between them.We can use this to generate our local label spread scheme,and we also experimented the co-occurrence phenomenon between labels,that is,if the user marks a tag,then another tag is likely to be marked;according to the above experiment,we implement the algorithm of local tag propagation to generate candidate tags.According to the interaction on the microblog between users,not only the forward propagation of the label is considered,but also the reverse propagation that may be brought about,and the different weight coefficients are given to them,thereby obtaining the local propagation of the label,and Generate candidate tags;next,we expand the candidate tags in a co-occurring way to get more candidate tags,and then remove some synonymous words by eliminating semantic redundancy,thus generating our final recommendation.Due to the extensiveness of the algorithm,we have dealt with two challenging problems in the traditional recommendation system,namely data sparsity and semantic redundancy.A large number of evaluation experiments have verified that our algorithm outperforms existing methods in recommending tag matching performance,and can obtain more accurate user tags.