搜索引擎点击模型综述

A survey of click models for Web browsing

搜索引擎用户在与搜索引擎的交互过程中反映出的隐性反馈信息(主要是点击行为信息)是搜索引擎用来改进结果排序的重要影响因素。然而,由于结果位置、展现形式等各种因素的影响,将反馈信息直接应用于搜索排序任务往往难以取得较好的效果。针对这一问题,研究人员提出了构建描述用户点击行为的点击模型,并基于不同的点击模型估计用户对展现结果的浏览概率,进而尝试去除结果展现位置等因素对用户行为的偏置性影响,以达到更好利用隐性反馈信息的目的。作为一种用户交互信息的有效利用方法,点击模型在学术界得到了充分关注,并在工业界得到了广泛的应用。本文是一篇针对点击模型发展过程的综述性文章,对点击模型发展过程中有代表性的多种模型进行了介绍。

The implicit feedback information contained in a user＇ s search interaction process makes an important contribution to the improvement of search ranking. However, since user behavior is affected by several factors （ or biases） caused by the ranked positions of the results, presentation styles, etc., it is difficult to directly adopt click information as a relevant feedback mechanism of the search sequence task. To shed light on this research question, researchers have proposed several click models to describe how users examine and click on results from the search engine result pages （SERPs）. Based on these models, it is possible to estimate the examination probability of search results and thus reduce the influence of behavior biases to obtain a justified estimation of the result＇ s relevance. Much attention has been paid to the click model in recent years because it helps commercial search engines to improve ranking performance. In this paper, recent efforts made in constructing click models were investigated and their differences were compared in both performance and application scenarios.