项目名称: 面向相关性反馈的搜索引擎用户点击模型研究
项目编号: No.61472206
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 计算机科学学科
项目作者: 马少平
作者单位: 清华大学
项目金额: 87万元
中文摘要: 结果排序技术是搜索引擎技术研究中的核心问题,而建立用户点击行为模型, 挖掘纷繁复杂的用户行为数据中蕴含的隐式相关性反馈信息则是这一技术问题的重要进展 方向。面对搜索结果中广泛存在的富媒体展现形式和多模态交互方式,以各搜索结果展现形 式相同为主要前提的同质性假设不再成立,这使得当前绝大多数点击模型在真实搜索应用环 境中受到越来越大的挑战。与传统点击模型构建方式不同,本项目提出应当基于海量规模用 户行为日志数据和眼动实验数据进行分析挖掘,对搜索引擎用户交互过程中客观存在的结果 展现形式、用户行为偏好和查询需求类型方面的异质特性进行深入分析与模型特征提取。在 此基础上,更加全面的描述用户点击行为,协助搜索引擎构建具有异质性描述能力的点击模 型,并借助机器学习方法实现对搜索结果相关性的估计,以更好的提升搜索引擎的结果排序 性能。
中文关键词: 搜索引擎;点击模型;用户行为分析;相关反馈
英文摘要: Search result ranking is one of the major concerns in search engine researches and click model construction which aims at improving ranking performance with the help of implicit relevance feedback information contained in click-through logs has been paid much attention. However, most existing click models assume that all search results should be homogeneous and are therefore not able to deal with search results in rich media formats or containing interaction functions. In contrast to the prevailing approaches, we proposed a different click model construction framework by taking the differences in result presentation, user preference and query information need into consideration. By collecting and analyzing both large-scale user behavior data and eye-tracking data, we plan to look into the practical information acquistion process of search users and extract behavior features to describe the heterogeneous nature of click-through behavior. After that, a learning based scheme would be adopted to provide relevance feedback results to improve ranking performance of commercial search engines.
英文关键词: Search Engine;Click Model;User Behavior Analysis;Relevance Feedback