The Design And Implementation Of Real-time News Recommendation System

The exponential explosion of information brings about information overloading problems and therefore generating category technology and search engine technology.However,the category technology simply covers the top category while search engine technology allows users to actively input keywords to search information.Consequently,personalized news recommendation system comes into being.Unitary algorithm could not make recommendations for users from many perspectives,which accordingly leads to the loss of recommendation result diversity.In order to improve the accuracy and diversity of recommendation,this thesis makes an analysis on existing recommendation algorithm and designs a mixed weighted news recommendation strategy in combination with traditional recommendation technologies.This thesis will make weighted combination on content-based recommendation algorithm and user-based collaborative filtering algorithm according to different weighted values and realize the complementary results to improve the accuracy of recommendation results and better make personalized news recommendation for users.This thesis takes news content modeling,user interest modeling and mixed algorithm modeling as the core contents in the recommendation system.As for news content modeling,this thesis firstly introduces relevant theories concerning news text pre-treatment.In view of the features of news contents,this thesis adopts linear Weighting method to retrieve key news words and classifies news with support vector machine.As for user interest modeling,this thesis analyzes users' preference for news browsing and finishes user interest modeling construction and upgrades through the collection of user behavior logs.As for mixed algorithm modeling,the content-based recommendation algorithm mainly determines news recommendation list through calculating the cosine similarity between news content vector and user interest vector.The user-based collaborative filtering algorithm could recommend similar news for users through establishing the user similarity matrix.Consequently,the results would undergo weighted combination according to different weighted values.In order to guarantee the accuracy of recommendation system,this thesis derives the weight ratio with optimal weighted effects through repeated practices.In addition,this thesis also sets up news time threshold to properly filter the recommendation results,which guarantees the timeliness of recommendation results to some extent.This thesis primarily determines the basic working contents through the introduction of system background significance and domestic and overseas research conditions.After making the detailed description about typical recommendation algorithm,this thesis analyzes system demands.In view of the large data scale and high user interest timeliness demands of news recommendation system,this thesis constructs this system,outlines the framework of mass real-time news recommendation system,relates the general design of recommendation system,elaborately observes system framework and key module realization process.This system offers more personalized and real-time information recommendation for users.