基于层次向量空间模型的用户兴趣表示及更新

Presentation and updation for user profile based on hierarchical vector space model

用户兴趣建模是个性化服务的基础与核心,而用户的兴趣会随着时间发生变化,这种用户兴趣漂移现象会导致系统预测用户兴趣的准确性下降.提出一种基于层次向量空间模型(VSM)的用户兴趣模型表示及更新处理机制,基于特征项形成兴趣主题,基于兴趣主题形成用户兴趣,由此建立层次型用户兴趣模型.采用基于用户浏览行为来计算用户对网页的兴趣度,快速估计网页兴趣度,以提高个性化系统的实用性,从而更好地满足用户个性化需求.实验结果表明,设计的用户模型表示及更新机制能有效提高个性化服务性能,准确率及召回率均有所提高.

The user＇s interest model is the basic and core component in a personalized services system, but user＇s interest will change over time, interest drifting problem cause the decrease of the prediction performance. Therefore, this paper presents a user interest model and the approach for dealing with the interest drifting problem based on a hierarchical vector space model, term form interest topic, user profile is formed with interest, then hierarchical vector space model （VSM） set up based on the user profile. Calculate the degree of user interest in the Web page based on the set of user browsing behaviors, to estimate user interest of Web pages quickly, in order to improve the practicality of personalized systems, which meet the personalized needs of users. The presentation of user model determines its ability to express user＇s real information and ability to calculate, but also it limit choice of the method for user modeling to some extent. In this paper, we view the text features as computable form of the instance and subject＇s characteristics, using term frequency-inverse document frequency（TF-IDF） formulation to calculate term weight, using cosine to calculate feature similarity. Although feature selection based on entropy can obtain better results, with high cost of computing and space consumption, which is not conductive to the online personalized recommendation system for practical application. As user profile need to update with the change of user＇s interests, the user model updation algorithm in connection with established model is proposed, which consider the updation of subject＇s characteristics and weight fully, using user feedback within a fixed window to update incrementally, with large overhead, and thus selecting the relevant parameters can improve the adaptability of the method. Experiment results show that the design of the user model and the update mechanism that can effectively improve the prediction performance for personalized services, with improvement on accuracy and recall.