| In today’s Internet era, people create a lot of information every day, the growthrate of infomation has been far greater than the upper limit of human beings. In thisear of informtion overload, it is crucial to filter information that people needed fromthe mass of information efficiently. From search engine to recommend systems. bothare designed to solve this problem. On the other hand, today’s internet era already isthe mobile internet era, more and more data is generated by user every day, and thedata’s value is also growing gradually. It is already the main trend to provide userspersonalized service based on their data. Among of all, microblog and other socialnetworks is a great source of user data. Using this data to analyse user’s interests, is aresearch hotspot in recent years.This paper is based on microblog data, research on how to mine user’s interestand its personalized recommendation. At first, we have a deep research and analysison LDA topic model in this paper, and study its application in the field of microblog,come to conclusion that the LDA model is not suitable for short microblog textdirectly. then we propose a user modeling method userd for microbolg’s user: mergeuser’s microblog set, use a user-topic-word three-layer model to indicate users’distributions of their topics and interest. Then, we propose a similar userrecommendation algorithm based on the similarity of users’ distribution of topics. Wedesign a set of experiments based on Vector Space Model and Hidden Markov model,and use the real microbolg data as data source. The experiments show that, ourmethod has a good effect. Secondly, in this paper, we make use of user’s social feature,propose a method to calculate user’s value based on its fans and concerns, we combinethe user’s value to our recommendation algorithm to show a better recommendationlist. At last, we also present a news recommendation algorithm based on the user’stopic model. We make a experiment and compare it to non-negative matrixfactorization, the result shows that, the proposed algorithm can not only find the user’sinterest, but can also get their interest distribution upon multiple topics. And in reality,users usually have multiple topics or interest, this algorithm can recommend usernews under more than one topic, and thus meet user’s requirement better. |
