基于用户兴趣变化和社会化标注信息的协同过滤推荐方法

随着Web 2.0技术的发展和推广,社会化标注系统为用户提供了有效表达自我和抒发感情的机会。针对社会化标签的特征,本文提出了一种考虑用户兴趣变化和用户标注信息的协同过滤推荐方法。首先,基于用户的历史记录信息构建了用户原始数据模型;然后将基于艾宾浩斯遗忘曲线的遗忘函数引入原始数据模型构建了用户兴趣模型,其中,利用融合时间权重的文本挖掘技术TF-IDF建立了基于项目类别标签的用户兴趣模型,综合标注标签加权频数和用户评分构建了基于标注标签的用户兴趣模型,融合时间权重和项目评分构建了基于评分的用户兴趣模型;进一步,基于用户兴趣模型并利用余弦相似性以及改进的Pcarson相关系数可计算融合用户评分和标签的用户兴趣相似度,根据用户兴趣相似度可为目标用户构建近邻集合从而生成推荐。实验结果表明,本文所提方法可以较好的考虑到用户的兴趣变化以及用户对标注标签的偏好,并通过对比实验证实该方法比传统的协同过滤方法推荐质量更高。

Collaborative Filtering Recommendation Method Based on User Interest Change and Social Tagging Information

With the development and promotion of Web 2.0,social tagging systems provide users with opportunities to effectively express their feelings.In view of the characteristics of social tags, a collaborative filtering recommendation method that takes into account user interest change and user annotation information is proposed in this paper.Firstly,a user original data model is built utilizing the user’s historical record information;and then a forgetting function based on the Ebbinghaus forgetting curve is introduced into the original data model to build the user interest model.Among them,the item category tag-based user interest model is established by using the text mining technology TF-IDF which incorporates the time weight;the social tagging-based user interest model is constructed by combining the user ratings and the weighted frequency of labeled tags;and the rating-based user interest model is built by integrating time weight and item rating.Further,based on the user interest model,the cosine similarity and the improved Pearson correlation coefficient,the user interest similarity combining user ratings and tags can be calculated.Based on the user interest similarity,a nearest neighbor set can be constructed for the target user to generate recommendation results.The experimental results show that the proposed method can take into account the changes of user’s interest and the user’s preference for tags,and the comparative experiments prove that this method has higher recommendation quality than the traditional collaborative filtering method.