Study And Implementation Of Personalized News Recommendation System Based On Topic Models

With the fast development of internet, browsing news on the internet becomes an important means of obtaining information. In the face of vast amount of information, more and more people hope to find their interested news and expect to receive pleased reading experience. How to rapidly acquire valuable information has become the crucial issue. In recent years, related research on personalized news recommendation system has received more and more attention and become an important research field.Based on the content-based recommendation method, this thesis conducted researches on news contents model representation, user interests’model and news recommend generation with exploiting novel topic model proposed in this thesis. The main work of this thesis includes following aspects.1. Analyzed personalized recommendation technology and research status, educed the issue of news contents model representation and user interests’mining in the personalized news recommendation methods.2. Proposed a content and topic characters based personalized recommendation scheme. This scheme is on the basis of extracting topic model and key words of the news to conduct character representation. Besides, users’preference for the topic and key words of the news as well as users’behaviour and scenario informations are taken into account to establish hierarchical interest model. On the basis of this model, taking into account the issue of too concentrated of the user interests, we proposed revised preference topic and set interest decay factor to represent and update interest move, discover long-term and short-term interests, reflect relationship between user interests and content of news accurately and completely. During the process of news recommend generation, we divided recommend list according to user interests model into several topic group and proposed a novel time-based recommend sorting calculation method. The diversity and novelty of news recommendation could be improved with this method, user interests adjusts to reading regularity much better. The goal of increasing recommend accuracy and effectiveness could be realized.3. Designed and implemented news recommendation system based on the news recommend scheme proposed in this thesis. Besides, with the test of system function, off-line simulation experiments were conducted to verify the improvement of news recommendation effect by adopting our scheme.