Research On Personalized Microblog Information Recommendation Based On LDA Model

The rapid development of the Internet has promoted the access and sharing of information based on the Internet.The activity of the netizen has made volume of data stored in the Internet grow explosively and the human society has entered the era of information explosion.Microblog as a typical representative of the social networking platform,has attracted large numbers of users to create and publish information in the platform because of its convenience in information dissemination,the diversity of content forms and the comprehensiveness of information coverage.The user can obtain the massive information resources in the platform.However,there are also some problems caused by information explosion such as the user can't obtain the information they are interested in quickly and accurately and many high-valued microblog information can't be effectively utilized.Personalized recommendation,which is based on user interests mining to recommend information to user contrapuntally,is an effective way to solve the above problems.Hence,this thesis mainly focuses on the research of user interests mining and scoring recommendation in personalized microblog information recommended.The main results are as follows:(1)When construct microblog user's interests model by directly using the LDA model,there are many problems like short microblog text length,semantic information lacks influencing the topic's modeling effect,fasiling to reflect the user's interests changing over time and others.Based on these problems mentioned above,this thesis has proposed an algorithm on account of text clustering and interests decay for microblog user interests mining(TCID-MUIM).By using the synonym merging strategy and the double single-pass incomplete clustering algorithm included in TCIDMUIM algorithm,the problem of short text length and lack of semantic information have been solved.And at the same time,by using the topic matrix compression method on account of time factors,the problem of user's interests changing over time has been solved.(2)In most of the existing recommendation methods,the recommend scores are usually obtained by calculating the similarity between topical probability distributions obtained after the topic's modeling.Therefore,during the grading process,these characteristics of microblog quality,freshness and others have neither been taken into account,nor the lexical probability distribution obtained after the topic's modeling been utilized.Then,a multi-angle personalized microblog recommendation algorithm based on user interest topics(MAMScore)has been proposed,hence,in this thesis,we use MAMScore to score microblog,and sorting and filtrating the Top-N microblog recommendation group which interested by user most.So that fresh microblog information with high quality can be recommended to the user under the premise of satisfying the user's interest.We perform experiments based on real microblog data set crawled from Sina microblog platform.The experimental results show that compared with the user interest topics mining by the traditional modeling method and the modeling method widely used in the microblog user interests modeling field by combining all user's historical microblog text into one document,the interest topics mined by the TCID-MUIM algorithm proposed in this thesis have a better discrimination,and fit the user's real preferences more.Based on the user's topic profiles,the MAMScore algorithm proposed in this thesis is more accurate in scoring and recommended microblog than those methods like scoring directly by the cosine similarity or the JS distance.