用户偏好建模：考虑长期与短期兴趣因素

Modeling User Preference Based on Long-term and Short-term Interest

摘要:

鉴于电子商务网站推荐系统的需要，将用户兴趣分为长期兴趣和短暂兴趣，并提出一种基于长期兴趣和短暂兴趣的用户偏好表示法。利用web服务器数据库的数据，采用无监督学习方法，对用户注册信息进行挖掘，提取出用户长期兴趣。基于向量映射，对web服务器日志上的用户使用记录数据和内容数据进行分析，提取用户短暂兴趣。通过用户反馈信息修正“粗糙”用户偏好文档，使得用户偏好文档更新得以实现。最后，应用了实证案例验证了该方法的合理性和有效性。

Abstract:

In view of the needs of e-commerce website for recommendation system, user interest is divided into long-term interest and short-term interest, furthermore, based on long-term interest and short-term interest, a way to describe user’s preferences is proposed. Utilizing the data from the web server database, using unsupervised learning, user’s registration information can be fully mined to abstract user’s long-term interest. Based on vector mapping, both the records data and content data on the server log is analyzed to abstract user’s short-term interest. Moreover, the rough profile presenting user’s preferences can be modified by dealing with user’s feedback, and that makes updating user’s preferences profile possible. Case analysis illustrates to a certain extent this method is reasonable and feasible.