Research On Personalized Recommendation System Of News Based On The Behavior Of Users

With the rapid development of Internet, offering personalized recommendations has been an essential service for each mainstream website. Offering various kinds of news by web portals is a traditional service on the Internet, but it still exists a big difference comparing with the booming e-commerce websites. Users may not go shopping online, but they are more likely to search news, which leads to that websites of information have a wide coverage of users. If the potential preferences of users can be mined and used to make recommendations, then would provide more value for the societies and economy.The main challenge to make recommendations for news is to provide articles to users which they may be interested in, thus the core work for recommendation systems is to analyze the historical reading behaviors of users, build corresponding modes and finally offer fresh news to users that they have not yet read.In this paper a novel news personalized recommendation system was studied, the key job is to recommend news to users effectively and accurately based on their historical data. The main works are described as follows.First, the algorithm based on the reading behaviors of users and the topics of news they read in the past was proposed. The similarities between users and news were computed separately based on the behavior data of users and the data of topics, which alleviated the cold-start problem.In the process of recommending news,the experiments showed that the behavior of users data plays a major role, which reveals that the performance of recommendation algorithms rely mainly on the behavior data of users.Second, another recommendation algorithm based on Hidden Markov Model was proposed. Based on users’ news reading data, a Markov data was built and used to predict which news one user would read. This algorithm was simple and did not need to extract textual topics of news, and besides it would filter some stale news without the consideration of time. At last the algorithm was improved by clustering users based on the idea that users who read similar users were similar. The improved algorithm proved to be better than the previous one in terms of performance.Third, a news recommendation system was designed based on the Hidden Markov Model. We analyzed the needs for news recommending, the recommending context such as users, websites and systems et al. was designed, which the recommendation system is the core, and the we designed the framework of recommendation system based on he Hidden Markov Model as well as the core of each module.Finally, we complete the display system of the algorithm results, and show the recommended results to the web interface.