面向社区的协同检索方法研究

项目名称： 面向社区的协同检索方法研究

项目编号： No.61202286

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 刘永利

作者单位： 河南理工大学

项目金额： 23万元

中文摘要： 随着社会化网络的飞速发展，协同检索成为信息检索领域的研究热点，它对于提高社区内用户检索的准确率和效率具有重要的实际意义。社区不断变化，因此需要对社区信息持续更新，但检索过程的特点为此造成了很大困难，包括：①数据稀疏；②特征空间维度高；③数据更新频繁。本课题围绕以上三个特点展开研究，内容包括：⑴三维空间相关性模型：建立由用户、查询和文档构成的三维空间，并采用概率方法量化三个维度间的相关性；⑵基于联合聚类的社区动态确定方法：针对检索过程的特点①和②，将原本仅用于分析二维列联表的信息论联合聚类方法进行扩展，使之适用于分析三维问题，进而动态确定用户社区；⑶增量学习机制：针对检索过程的特点③，从数据所在维入手，增量更新三维概率关系及联合聚类结果。本课题基于三维空间的概率关系，重点围绕社区的动态更新问题，兼顾理论分析和实践验证，为协同检索方法的进一步研究与应用提供新的思路。

中文关键词： 信息瓶颈；信息检索；模糊聚类；联合聚类；增量聚类

英文摘要： With the rapid development of social network, collaborative retrieval has become a research focus in Information Retrieval field, which has important practical significance to raising the accuracy and efficiency of Web search in communities. A community is constantly changing, which makes it obligatory to identify communities in dynamic networks, but such characteristics of search process will become bottlenecks, as ①sparsity, ②high-dimensionality and ③dynamic data. This topic mainly includes following three parts. ⑴Probability methods are employed to express the relevance amongst user, query and document. ⑵When dealing sparse and high-dimensional data, we innovatively extend information-theoretic co-clustering methods originally just used to analyze two-dimensional contingency tables to be is suitable for three-dimensional data. ⑶For dynamic data, we start with the dimensionality that needs to be updated, and incrementally renew the three-dimensional relevance and co-clustering results, as can improve the update efficiency. We also analyze and discuss the recommendation method based on collaboration and real-time property. Our topic is based on probabilistic relevance of the three-dimensional space, focuses on identifying communities dynamically, takes into account both theoretical analysis and experimental ver

英文关键词： information bottleneck；information retrieval；fuzzy clustering；co-clustering；incremental clustering