目前,社会化标注已经成为个性化信息推荐领域中的研究热点之一,标签质量对于推荐效果的影响也受到了广泛关注.本文针对标签的质量问题,指出用户标注偏差普遍存在于标注系统中,尤其是形式偏差,给用户兴趣模型的合理提取形成了阻碍.基于此,我们提出了主流标签的概念,以其体现的大众智慧来克服标注偏差所带来的影响,通过分析资源中标签的平均标注率进行主流标签数量的确定,实现资源模型和用户协同模型的构建,并进一步结合兴趣度对用户协同模型加以了改进.最后,基于Delicious的数据和用户参与评分法,文章运用余弦相似性对模型推荐效果进行了验证.%Social tagging is becoming a hot issue in personalized information recommendation recently, and theinfluence of tag quality to information recommendation is also emphasized. This paper first address the problem of tagging deviant which is found widely existing in tagging system, especially the form deviant make a reasonable user profile difficult to construct . In order to overcome the shortages triggered by tagging deviant, a concept of representative tag is put forward. Through the analysis of average tagging rate of each tag in typical resources, the number of representative tag is determined. Then, resource and user profile models are constructed based on representative tags and interest degree, and recommendation algorithms carried out by cosine similarity are presented as well. Finally, the effect of our recommendation model is tested based on Delicious data and user evaluation.
展开▼
