基于分布式语义的融合式社会化标注语义分析方法研究

As one of the killer applications of social computing, social tagging has become more and more popular in recent years. However, as the key issue in social tagging research, for a long time, semantic ambiguity has been the major obstacle to developing social tagging as the reliable efficient way of content-based information retrieval. This project is dedicated to develop systematic way of semantic analysis for social tagging which is capable of solving semantic ambiguity problem effectively. Thus, all parts of the project are designed for solving the semantic ambiguity problem in social tagging systems. In detail, with the support of distributional semantics based on second order co-occurrence distribution, the project consists of 4 parts. Firstly, the method of mapping tag to concept between folksonomy and ontology is developed. Then, on the basis of the first part, the way of detecting synonymy implied among tags of folksonomy is explored. As the third part of the project, the semantic analysis method for ambiguous tags is proposed based on the preceding research achievements. Finally, the folksonomy oriented ontology evolution is studied and the social tagging semantic analysis system based on the fusion of folksonomy and ontology is designed and implemented for empirical evaluating the achievement of the project. The accomplishment of the project is valuable for promoting the value of social tagging in the field of information retrieval. Besides, the achievement of the project can play important role for improving the research on information retrieval and/or recommendation based on social tagging.

社会化标注是近年流行的重要的社会计算应用方式。然而，语义模糊性阻碍着社会化标注成为高效且可信赖的基于内容的信息检索方式，是社会化标注研究领域的核心问题。本项目围绕社会化标注系统中的语义模糊问题开展研究，旨在形成系统化的、可有效处理语义模糊问题的社会化标注语义分析方法。具体而言，以基于二阶共现分布的分布式语义分析为核心方法，首先解决大众分类标签与本体概念的匹配问题；然后，在此基础上，研究大众分类中标签同义关系识别方法；接着，在前两项研究基础上研究多义标签语义分析方法；最后，研究面向大众分类的本体演化方法，开发融合大众分类与本体的社会化标注语义分析系统，以实证研究方式验证本项目的研究成果。本项目研究的成功实施可提升社会化标注系统在整个信息检索领域中的应用价值，对应的研究成果可为深入完善基于大众分类的信息检索及推荐方法提供直接的支持，对本体领域的研究具有参考价值。

由社会化标注系统形成的大众分类在个性化推荐领域和信息检索领域已经得到了广泛的应用。社会化标注系统的成功主要缘于用户可以随意使用标签标注资源。然而，正是这种不规范的标注方式使得社会化标注系统及大众分类长期受到语义模糊问题的困扰阻碍着社会化标注系统进一步发展。针对这些问题，我们构建了基于用户兴趣的关系网络，并在该网络上通过社区发现算法进行用户社区发现，识别多义词。应用二阶共现分布构建分布语义模型。研究指标体系，筛选数据并实现了数据的可视化。提出了处理动态数据的框架和高噪声环境下数据特征的选择算法。最后，开发了基于 RDF 的学术资源标注推荐系统.