Research On Key Techniques Of Information Recommendation On Social Media

As the representative of the social media,Microblog has gradually become a new carrier of information dissemination and information sharing platform.The widespread of social media leads to the information explosion.It is difficult for users to get real useful information and therefore leads to information overload and content fragmentation problems.All of these problems bring opportunities to the research of information recommendation on social media.Social media contains a large amount of user generated content.Information recommendation on social media aims to infer user preferences and provide user with information of interest through understanding and analyzing user generated content.Specifically,information recommendation on social media contains three key elements.The first element is the information,which is the object of information recommendation.The content of information itself and the relationship between information provide a basis for quantifying the user's interests.The second element is the user,which is the subject of information recommendation.Mining user's interests and preferences can help to carry out more effective recommendation.Finally,the recommendation algorithm links information to users,is very important to information recommendation on social media.The research work of this paper covers the above three key elements.Using microblog as a research platform,we explore the key technologies of information recommendation on social media,making full use of users' text data and collective wisdom of users,combined with techniques of deep learning and topic modeling.The main contents of this thesis can be summarized as follows:1.This thesis proposes a novel method based on the topical attention mechanism to recommend the hashtags of microblog,to solve the problem of the relevance of the recommended information.Hashtags can organize microblogs with similar content or similar topics in the form of a topic or keyword.Aiming at the differences in the word usage of post and hashtags in microblog and the feature sparsity problem,we model the problem of hashtag recommendation as a classification problem and adopt LSTM to learn the representation of a microblog post.In order to highlight the role of microblog topic,we propose a novel attention-based LSTM model which incorporates topic modeling into the LSTM architecture through an attention mechanism.Experimental results on hashtag acquisition show that our model significantly outperforms various competitive baseline methods.Furthermore,the incorporation of topical attention mechanism gives more than7.4% improvement in F1 score compared with standard LSTM method.This lays a good foundation for the subsequent research of microtopic recommendation.2.This thesis proposes a method to predict the popularity of microblog based on the similarity of the influencing factors,to the problem of the importance of the recommended information.For information recommendation,not only the relevance of information content,but also the value of the information itself needs to be paid attention to.Thus we can provide microblogs with high popularity from the vast amount of microblog information to the users.Noting that the publishers of micro-blog play a decisive role in the popularity of microblog,this paper proposes a user centered method to predict the popularity of microblogs based on the similarity of influencing factors.The main idea of this method is to use the user's historical microblog information and rank the microblogs according to content and time similarity.Then we choose the most similar K candidates to the current microblog of the same user,and use the weighted sum popularity to predict its popularity.This method is simple and effective,and can be incorporated into the existing model as a feature to further enhance the prediction results.3.This thesis proposes a supervised topic model to model user interest,to achieve information complementary cross platform and solve the problem of completeness of user information.In this paper,we take the consumption interest as the research object,using user's content in social media to infer user's consumption interest categories.To solve the difficulties of cross platform data acquisition,we propose to mine the linking traces across social media and e-commerce website and therefore identify the same users cross platforms.We propose a novel supervised topic model naturally links users' published content and following relations on microblogs with their consumption behaviors on ecommerce websites.Experimental results on microblog and Jingdong dataset show our model outperforms the state-of-the-art methods in the task of user consumption interest modeling.Our model can also automatically learn meaningful consumption-specific topics and words which can be used for other downstream applications.4.This thesis proposes a microtopic recommendation model with rich information incorporated,which can alleviate the cold start problem of information recommendation and improve the recommendation results.In this paper,a hybrid recommendation model is proposed,which combines user adoption behavior,user microtopic text information and attribute information.The model is based on the topic model and collaborative filtering model,which is combined with the text information to build a good explanation for the relationship between the user and the microtopic.On the one hand,the user's preference is studied by collaborative filtering technology.On the other hand,the user's interest is grasped by the topic model.The experimental results show that the proposed method can significantly improve the performance compared with the state-of-the-art hybrid recommendation methods,especially in the cold start cases.In Summary,on the one hand,in this paper we dedicated the basic problem of information recommendation on social media,including information modeling and user modeling.On the other hand,from the perspective of information recommendation algorithm,we put forward the information recommendation model with rich information on social media.This study has made some preliminary results,with the development of natural language understanding and information recommendation technology,we believe that the research of information recommendation in social media will make a greater breakthrough in the future.At the same time,the research of information recommendation in social media will promote the development of other related research.