基于眼动和主题模型的个性化实时查询扩展模型的研究

发布时间：2018-07-06 11:36

本文选题：查询扩展 + 眼动(Eye　；参考：《天津大学》2016年硕士论文

【摘要】：对于大部分用户甚至是有经验的用户来说如何形成一个较好的查询能够获得更好的搜索结果仍然被认为是信息检索(Information Retrieval)的一大难题。查询扩展往往是提高检索性能的有效方法。通过找出语义上与原始查询比较相关的词语、概念等,再结合用户的原始查询,使得扩展之后的查询能够提供更多的积极信息来从海量信息中找出与用户查询相关的文档,改善用户搜索体验。传统的查询扩展技术已经在很大程度上解决了查全率(Recall)低下的问题,但是对于查准率(Precision)上却很难去的较令人满意的结果。个性化的查询扩展部分解决了查准率较低的问题。但是传统的个性化的查询扩展往往利用用户过去的搜索数据而且很难捕捉用户在本次查询中的需求动态变化,很难实时地根据用户与搜索引擎的交互来满足用户的查询需求。眼动(Eye Movements)能够在不引起用户注意的情况下实时地捕捉到用户的注视信息,进而提供用户的实时搜索行为数据,被视为用户研究和个性化的搜索的一个全新的方向。因此,若能将眼动技术应用在当前亟待解决的个性化的查询扩展上来,将是一个全新的启发式的研究方向,具有重大意义。论文的主要研究工作分为以下几个方面:第一,对眼动(Eye Movements)在IR上的主要应用进行了概述。除了介绍眼动在IR上的应用之外,着重介绍了如何利用眼动(Eye Movements)实时捕捉用户的动态搜索数据以及如何利用捕捉之后的眼动数据来进行个性化的查询扩展。第二,介绍了主题模型与眼动(Eye Movements)的结合方法。仅仅利用捕捉到的用户的眼动(Eye Movements)数据进行个性化查询扩展词的计算,还不能够充分挖掘用户的潜在搜索意图,为此利用主题模型Latent Dirichlet Allocation(LDA)来发掘和用户查询潜在相关的查询词,提高检索成绩。第三,建立实时查询扩展模型(Real-Time Query Expansion,RTQE)。通过创新性地结合眼动和LDA,该模型能够在用户点击若干篇文档之后,记录用户的注视数据,在用户刷新当前搜索结果界面或者点击下一页的同时根据用户若干分钟前的注视数据通过建立的RTQE模型重新对已有的搜索结果进行排序和优化,提升用户体验。
[Abstract]:For most users and even experienced users, how to form a better query to obtain better search results is still considered a big problem in Information Retrieval. Query expansion is often an effective way to improve retrieval performance. By finding out the words, concepts and so on, which are related to the original query semantically, and combining with the original query of the user, the extended query can provide more positive information to find the documents related to the user query from the massive information. Improve the user search experience. The traditional query expansion technique has solved the problem of low recall to a great extent, but it is difficult to get satisfactory results for precision. The personalized query extension solves the problem of low precision. But the traditional personalized query expansion often makes use of the user's past search data and it is difficult to capture the dynamic changes of the user's demand in this query. It is difficult to satisfy the user's query demand according to the interaction between the user and the search engine in real time. Eye movements (Eye-Movements), which can capture the user's gaze information in real time without attracting the user's attention, and then provide the real-time search behavior data of the user, are regarded as a new direction of user research and personalized search. Therefore, if the eye movement technology can be extended to the current personalized query, it will be a new heuristic research direction and has great significance. The main research work is as follows: first, the main applications of Eye movements in IR are summarized. In addition to introducing the application of eye movement in IR, this paper mainly introduces how to use eye movements to capture the user's dynamic search data in real time and how to use the captured eye movement data to expand the query. Secondly, the method of combining theme model with Eye movements is introduced. Only using the captured user's eye movements to compute the extended words of personalized query can not fully excavate the potential search intention of the user. Therefore, the topic model named Latent Dirichlet Allocation (LDA) is used to discover the query terms related to the user's query potential. Improve retrieval results. Thirdly, real-time query expansion model (RTQE) is established. With an innovative combination of eye movements and LDAs, the model can record the user's gaze data after a user clicks several documents. At the same time the user refreshes the current search results interface or clicks the next page and reorders and optimizes the existing search results according to the user's gaze data several minutes ago through the RTQE model to improve the user experience.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【参考文献】