个性化信息服务的多维多粒度用户兴趣建模研究

项目名称： 个性化信息服务的多维多粒度用户兴趣建模研究

项目编号： No.61262033

项目类型： 地区科学基金项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 陈辉

作者单位： 江西财经大学

项目金额： 45万元

中文摘要： 用户兴趣建模是当前Web信息检索研究的热点，根据用户兴趣提供个性化服务是解决信息过载和信息迷航的根本途径，但现有兴趣模型普遍存在信息描述不充分、视角单一和集成度低等问题。项目采用多维空间模型描述兴趣多方面特征，并对时间与主题属性进行多粒度描述，建立多维多粒度用户兴趣模型UIM3。在此框构下，研究兴趣时间与主题最小粒度选择、多粒度时间索引树多层聚合、兴趣主题图动态聚合与分解等策略，动态维护不断变化的用户兴趣；研究滑动窗口大小动态调整策略和兴趣权值衰减机制，设计兴趣漂移实时检测算法，捕获兴趣变化；研究基于隐马尔科夫模型的兴趣预测模型，设计相关预测算法；研究多维多粒度用户兴趣模型性能评估体系，构建基于灰度聚类的性能评估模型，综合评测UIM3模型性能。项目将构建一个集成兴趣实时跟踪、漂移检测和兴趣预测的综合模型，实现相关策略和算法，可为解决信息过载和信息迷航问题提供理论支持，提高个性化服务质量。

中文关键词： 用户兴趣模型；概念主题抽取；相关性度量；数据挖掘算法；多维度多粒度

英文摘要： At present, the study on user interest model has become one of the hotspots of the research on Web information retrieval. The foundational way to solve the problems of information overload and being lost in internet lies in providing personalized services for the user according to their interests.However, the existing interest models have ubiquitous problems, such as insufficient information, unitary perspective and low integrated level. In this project, we describe the interests in a multiple dimensions space model according to the attributes of the interests, in which the time and theme of interest are represented at multiple granularities. Based on the space model, we build a user interest model named UIM3 at multiple dimensions and multiple granularities. Under the framework of UIM3 model, we study a series of strategies to dynamically maintain the UIM3 model and real-time tracking the time varying interests, including the strategy of choosing the minimal granularities for interest's time and theme, the strategy of dynamically aggregating the nodes in multi-granularities time-index-tree, and the strategy of dynamically aggregating or splitting the vertexes in multi-granularities theme graph. Also, we study the strategy of dynamically adjusting the size of sliding window and the mechanism of gradually fading

英文关键词： User interest model；Concept topic extraction；Similarity measurement；data mining algorithm；Multiple dimensions and multiple granularities