Research Of User Multi-Interest Modeling Based On Semantic Similar Network In Personalized Information Retrieve

This paper proposed an improved Vector Space Model representation of user single-interest together with its dynamic learning algorithm. Experiment shows that the algorithm can catch and record user's latest interest in time, and can self-adaptively adjust and update user profile. On this basis, taking into account the diversity of user interest as well as the synonymity and semantic correlation of word expression, the paper further proposed a method for user multi-interest modeling based on semantic similar network (SSN), which uses SSN to expand user feature words with its synonyms and correlative words on the knowledge level, and divides user interest into multi-categories, thus to establish a multi-interest user model. In the personalized recommending test, the system taking the new method has better effect than that taking the improved Vector Space Model representation. Besides, the paper studied feature selection in user modeling and proposed a feature selection method combining term frequency and TFIDF according to part-of-speech tagging. It shows in experiment that the effect of combining method based on part-of-speech tagging is better than that of using TF or TFIDF separately. In the meantime of theoretical study, the paper presented an application of the single-interest and multi-interest user modeling technology in personalized information retrieve, which is an Intelligent TV Program Recommending System. All tests based on the system prove the validity and high precision of feature selection method and dynamic learning algorithm. They also indicate that the user multi-interest modeling method based on SSN has good modeling performance.