Research On Interactive Personalized Recommendation Technique Based On Tag Feature

With the rapid development of Internet,the problem of “Information Overload” is increasingly severe,and people are facing the crucial issue about how to find the demanded contents from the massive information in Internet.The personalized recommendation system is the common approach to deal with the Information Overload problem,and its successful application in e-commercial area has attracted a wild range of attention,as well as the interests of application in other areas.The personalized recommendation system analyzes the preference of user interest by collecting the information of user behaviors,and then makes the personalized recommendation to different users.However,there are some problems in the personalized recommendation algorithm.The performance and effectiveness of the recommendation algorithm will decrease significantly as the increase in the number of alternative resources;“Cold start” is one of the common problems in recommendation system.This thesis focuses on the above problems,and the main contents include:1)In order to quickly locate the demanded resource item and rapidly reduce the alternative resources,this work proposed the interactive recommendation model based on tag.In each interaction round with the user,the system provides certain tag attributes for users to choose,which can divide the resource set with most distinguishing information,so as to optimize the resource set partition.2)“Cold start” is one of the common problems in recommendation system.This work deals with the problem by vote ranking algorithm.If the user information is insufficient,the system makes use of other users' selection and marking to rank the resource items,and recommend the best items to the users for their feedbacks.3)With the interactive recommendation model,this work proposed the improved item-based collaborative filtering recommendation algorithm,which uses time sliding window algorithm to build user interest model and extract the tag features.Then this work find the similar resource items according to the tag feature vectors of different items.Finally,the improved item-based collaborative filtering algorithm is realized by estimating the users' preference.In this thesis,we proposed an improved solution for "cold start" problem and an improved item-based collaborative filtering recommendation algorithm based on interactive recommendation model.This work illustrated the effectiveness of the interactive personalized recommendation algorithm by the results of comparison experiments and the performance in terms of precision and recall rates.Especially,the interaction tests discussed the relationship between the interaction round and algorithm performance.