User Interest Analysis And Personalized Information Recommendation Based On Microblog

In the past ten years, the information on the Internet is growing rapid?ly. We walk from the lack of information era into the era of informationexplosion. The way people getting information is also transforming, frommanual searching information to search engine, and now the recommendersystem. One of the most important steps to recommend useful informationto user is how to get the interest of users effectively. The rise of micro?blog gives us a new and huge data pool to analysis user interest. It rapidlybecomes a research hotspot in recent years.In this thesis, we talk about and explore the approach to analysis userinterest from micro-blog data, and how to make personalized recommenda?tion. Compared with the existing works, this thesis has several differences.First, considering about the character that the content of each micro-blogis very short, we construct the topic model on external knowledge dataset,instead of on the micro-blog data directly, so as to enrich the semantics ofmicro-blog content. At the same time, this approach avoids the problemof determineing the number of topics on micro-blog data. Second, we ob?served that not all the posts are related t’o users interest. Some posts are theso called ’noisy posts. These noisy posts have an bad impact on the result.So we talk about the features used to recognize noisy posts from differentaspects, and then construct a combined classiifer to filter the noisy posts. At last, we believe that user’s interest will change over time, so we argue to use the time weighted topic distribution to represent user’s interest. In our experiments, we compared our algorithm with two other approaches, that’s the non-negative matrix factorization approach and apply topic model on micro-blog data directly approach. The results show that our approach can mine user’s interest effectively. What’s more, our approach performs bet-ter when there are less posts under a user, or when there are many noisy posts under a user.