Abstract: To solve the disordered annotation problem occurred in classification mark, this paper proposes an algorithm supporting for dynamic ranking of Web music tag. The efficacy of classification tags is calculated based on the log of click tags and other operation data. On the basis of fuzzy mathematics, this paper introduces respectively the tag efficacy measurement model, efficacy metric algorithm and efficacy drift measurement algorithm. Firstly, this paper constructs a hierarchical structure model for music tag system according to the metadata attributes of Web music. Secondly, this paper designs a group of classification tag efficacy indexes, and computes the efficacy using the fuzzy mathematics theory with adaptive characteristics. Thirdly, this paper proposes the efficacy models of classification tag respectively on the long-term and short-term, and defines efficacy volatility index to measure the stability of long-term and short-term efficacy. Finally, this paper employs a case study of music platform based on practical data to verify the efficiency of the proposed algorithm and show its applicability.

Key words: group behavior, tag efficacy, efficacy drift

摘要： 为改善多分类标注时的无序标注问题，提出了一种适用于网络音乐标签动态排序的支持算法。基于点击标签和其他操作的日志数据计算分类标签效用度，以模糊数学理论为基础，提出了标签效用度量模型、效用度量算法与效用漂移度量算法。根据网络音乐的元数据结构属性，构建了音乐标签体系的层次结构模型；设计了一组能够表达分类标签效用的特征指标，并利用带有自适应特性的模糊数学理论计算标签效用度；在分类标签效用模型的基础上提出了长期和短期效用漂移的模型，并创新地用效用波动度指标来衡量长、短期效用的稳定性；在音乐平台案例实际数据的基础上进行了算法验证，表明了该算法的适用性。

关键词: 群体行为, 标签效用, 效用飘移