社会化标注中用户动态兴趣主题挖掘

发布时间：2018-01-09 06:26

本文关键词：社会化标注中用户动态兴趣主题挖掘　出处：《浙江理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：在社会化标注中,用户采用标签的方式对自己感兴趣的资源基于自身理解进行无约束标注,标签作为用户所选择的关键词,反映了用户对信息资源的看法及其兴趣所在。用户兴趣的发现获取是个性化推荐的基础和核心,现有研究主要聚焦于从数量和结构上考虑用户与标签的关系,而实际情况往往是,标签只是反映用户兴趣的一个单一词汇,用户的兴趣往往是由多个标签形成的集合所刻画的主题,而且标签随用户标注时序动态变化的特点也会影响用户兴趣的刻画。因此本文结合国家自然科学基金项目“泛在计算环境中社会化驱动的情境感知个性化信息服务研究”(项目编号:71471165),研究了社会化标注中用户动态兴趣主题的挖掘,系统地阐述了用户兴趣主题模型的构建方法,提出了基于动态标签关联指标来挖掘有效的用户兴趣主题并展开实验研究。本文的主要工作和贡献如下:(1)构建了主题模型获取标签主题。以“用户-标签”二元关系为基础,把用户标注信息作为语料库,标签被看作是这个语料库中的词,从而构建LDA模型挖掘潜在的标签主题。将原本杂乱无章、动态变化的标签划分成具有主题的簇,从而克服标签本身的不确定性和不一致性。(2)提出了动态关联指标获取用户标签兴趣。建立关联空间模型来描述用户与标签集的联系,通过分析社会化标签系统的具有时序特征的标注过程,考虑标签的时序特征,定义了反映数量特征的强度指标和反映时间变化的稳定性指标,通过动态关联指标获取用户动态标签兴趣。(3)用户动态兴趣主题挖掘。结合标签主题模型和用户动态标签兴趣,将两者进行相似性计算,得到用户动态兴趣主题,从而有效地实现了用户标签集到用户兴趣主题的动态对应。(4)选取来自Last.fm社会化标注平台中真实的用户标注数据,开展用户兴趣主题挖掘的实验研究,利用覆盖度Acc和准确率Rec对本文提出的基于动态关联指标挖掘的用户兴趣主题模型的有效性进行验证。研究表明,本文提出的基于动态关联指标挖掘的用户兴趣主题在有效性方面有较好表现,且明显优于基于TF方法和基于TF-IDF方法挖掘的用户兴趣主题,研究结果对于个性化推荐具有较高的实践应用价值。
[Abstract]:In social tagging, users use tags to label the resources they are interested in based on their own understanding. Tags are the keywords chosen by users. The discovery of user's interest is the basis and core of personalized recommendation. The existing research mainly focuses on the relationship between user and label in terms of quantity and structure. However, the actual situation is that tags are only a single word that reflects the interests of the user, and the interest of the user is often described by a collection of multiple tags. Moreover, the characteristics of label dynamic change with user tagging time series will also affect the description of user interest. Therefore, this paper combines with the National Natural Science Foundation of China. " Research on social-driven context-aware personalized information service in pervasive computing environment "(. Item No.: 71471165). This paper studies the mining of user's dynamic topic of interest in social tagging, and systematically expounds the method of constructing user's topic of interest model. In this paper, we propose a dynamic tag association index to mine effective user interest topics and carry out experimental research. The main work and contributions of this paper are as follows: 1). The topic model is constructed to get the label theme, which is based on the "user-tag" binary relationship. The user tagged information is regarded as the corpus, and the tag is regarded as the word in the corpus, thus the LDA model is constructed to mine the potential tag topics. Dynamically changing labels are divided into clusters with themes. In order to overcome the uncertainty and inconsistency of label itself, the dynamic association index is proposed to get user tag interest. The association space model is established to describe the relationship between user and tag set. By analyzing the tagging process of the social label system with time series characteristics and considering the time sequence characteristics of the label, the strength index reflecting the quantitative feature and the stability index reflecting the change of time are defined. Through dynamic association index to obtain user dynamic tag interest. 3) user dynamic interest topic mining. Combining tag topic model and user dynamic label interest, the similarity between them is calculated. Get user dynamic topic of interest, thus effectively realize the user tag set to user topic of interest dynamic correspondence. 4) select the real user tagging data from Last.fm social tagging platform. The experimental research of user topic of interest mining is carried out, and the validity of user topic of interest model based on dynamic association index mining is verified by using coverage Acc and accurate rate Rec. The topic of user interest based on dynamic association index mining presented in this paper has a good performance in terms of effectiveness, and is obviously better than that based on TF method and TF-IDF method mining user topic of interest. The research results have high practical value for personalized recommendation.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1;F49

【相似文献】