Research And Implementation Of Key Techniques In Personalized News Recommendation System

The appearance and popularity of the Internet gives people a lot of information. With the rapid development of network, the information has grown substantially which brings the information overload problem. Internet users often feel powerless in facing with such a large amount of news. They do not know how to acquire the news which they actually concerned effectively in avoidance of spending a lot of time and effort finding these news. In the area of Chinese news, as hundreds of thousands news appearing every day, users often cannot find the news which they care about quickly.Given this, researchers have proposed the use of recommendation system for the personalized recommendation of Chinese news. The recommendation system can recommend the user’s interests based on his profile and interest. The system will find a user’s interest and guide the user to find his own information requirements which improves retrieval time and reduces cost in energy.In the research topic, we improve three key technical points of a good personalized news recommendation system which respectively deal with three difficulty containing news text clustering, news online classification and user behavior modeling. These methods have been successfully used in the Sohu news recommendation system currently. Firstly, when dealing with news text clustering, this paper puts forward a new method in which neural network language model is first applied to text clustering. Text clustering is first converted to its dual problem called word clustering. With neural network language model, we can get word vector which can be used in the fuzzy k-means of the Chinese news keyword set. Based on the keyword clustering result, we can get text clustering result of Chinese news by a single transition. Experiments have show this method’s running speed is five times faster than LDA. Secondly, we propose a news online classification method by comparing the relationship between keywords and the feature words. Finally, we use user keywords to show users’ interests and to model users.