The Research And Implementation Of Organization Generated Tag Based Recommendation Algorithm With User's Tag Preference Habit

With the development of network technology, tags have been widely used to enhance the effectiveness of recommend systems. The existing researches have focused on the usage of the user's free tags. However, they have ignored the resource tags which were created by the companies or organizations while producing the resource. The number of free tags is huge, but there are sparse problems and ambiguity existing in them. While the number of resource tags is small, the meanings are more accurate. How to use the resource tags to enhance the recommendation effect is the target of this research.In this thesis, the existing tags were divided into two categories: the user tags which were generated by the user freely, and the professional and rigorous resource tags which were generated by the organization. Detailed analyses and comparison works have been done with these two kinds of tags, and researches of recommendation technology were conducted towards the information of these two kinds of tags. In this thesis, a novel collaborative filtering recommendation model, TRUP(Organization Generated Tag Based Recommendation With User's Tag Preference), was proposed to optimize the recommendation algorithm by using tags. Making use of the characteristics of the resource tags, the model graded the tags through converting the scores rated to the items by user. Then the model established the user's tag preference and calculated the user similarity matrix. In addition, in order to solve the problem of low computational performance of TRUP model in large amount of data, a solution was proposed to improve and optimize the relational matrix operations of the model by using K nearest neighbor algorithm. And the solution reduced the time complexity of the model and optimized the overall recommended effect of it. Finally, the comparison of the performance of TRUP model with other commonly used recommendation algorithms in three types of indicators showed that the TRUP model was feasible and effective.In this thesis, it was found that the time decay algorithm cannot be effectively applied to the TRUP model when time information was tried to be used. Through analyzing the user's behavior, the concept of user's tag preference stability was presented, the experimental algorithm was designed and the result of experiment was verified. It was proved that the user's tag preference is stable.