基于查询偏好的个性化搜索引擎的研究与实现

发布时间：2018-02-24 08:01

本文关键词： 搜索引擎查询推荐朴素贝叶斯点击预测相似计算　出处：《江南大学》2016年硕士论文　论文类型：学位论文

【摘要】：如今,高速发展的因特网为用户的生活提供便捷的同时,也不可避免的出现让用户难以获得所需信息的烦恼。对此,搜索引擎的产生为用户查询信息的方式带来了极大改变。作为一种常用的网络信息检索方式,搜索引擎已经发展为每个用户上网都离不开的重要工具,它的出现得到人们的广泛关注和使用。然而,传统信息检索工具也存在诸多不足之处。主要因为传统搜索引擎采用关键词匹配的全文检索方法,缺乏有效的对用户本身个性化信息的挖掘,从而不能有针对性的为用户提供个性化查询服务。所以在考虑和分析不同用户的信息需求的基础上,个性化搜索系统应运而生,它的出现能够在一定程度上满足用户的个性化信息需求。本文借鉴了电子商务网站中广泛采用的信息推荐技术,提出了较为可行的个性化信息推荐方案,即在用户搜索中使用查询推荐技术,以实现个性化搜索推荐。相关研究表明,用户的查询点击历史反应了用户的搜索习惯和查询偏好,所以本文对用户的历史查询数据进行深入分析,提出了用户点击模型,预测用户查询相关性并最终给出推荐。文中的研究主要集中在查询推荐上,首先根据用户的搜索点击历史数据,使用朴素贝叶斯理论训练出一个用户点击模型,根据这个点击模型对用户当前提交的查询进行分析,预测查询与链接的点击率,然后根据反向点击图模型将预测的值分配给相应查询,据此对查询进行相关度排序,将前k个最高预测值的查询推荐给用户。其次,本文在单个用户历史数据的基础上,对其进行补充,提出协同相似计算的用户查询推荐技术,将具有相同搜索行为的用户数据汇集到目标用户。该方法可以解决目标用户数据量不足的情况,另外还可以为用户提供一定的查询推荐的扩展性和新颖性。在用户的相似计算中,将每个用户的历史查询日志整合为一篇文档,使用余弦向量模型计算文档间的相似性；再使用推荐系统中的协同推荐理论,将用户对每个链接的点击频率比作为相关偏好评分,基于这些点击评分,使用改进后的欧氏距离计算用户间的相似度；最后将两种相似计算方法采用线性加权的方式整合,计算出目标用户的相似用户集,最终将采用点击模型进行查询推荐。本文最后基于查询推荐的分析,实现了一个简易的个性化搜索引擎系统。在本系统中,将相关查询推荐算法引入到了系统中,并将用户点击模型应用到网页排序上,实现了个性化的查询推荐和网页排序功能。
[Abstract]:Today, the rapid development of the Internet not only provides convenience for users, but also inevitably makes it difficult for users to obtain the information they need. The generation of search engines has brought great changes to the way users query information. As a common way of information retrieval, search engine has developed into an important tool that every user can not access the Internet without. However, traditional information retrieval tools also have many shortcomings, mainly because traditional search engines use keyword matching full-text retrieval methods. There is a lack of effective mining of personalized information for users, so it can not provide personalized query services for users. Therefore, on the basis of considering and analyzing the information needs of different users, the personalized search system emerges as the times require. It can meet the individual information needs of users to some extent. This paper draws lessons from the widely used information recommendation technology in e-commerce websites, and puts forward a more feasible personalized information recommendation scheme. In other words, query recommendation technology is used in user search to realize personalized search recommendation. Related research shows that the user's query click history reflects the user's search habit and preference. So this paper deeply analyzes the historical query data of users, puts forward the user click model, predicts the relevance of user query and finally gives the recommendation. The research in this paper is mainly focused on query recommendation. First of all, a user click model is trained by using naive Bayes theory according to the user's search click history data. According to this click model, the query submitted by the user is analyzed, and the click rate of the query and link is predicted. Then, according to the reverse click graph model, the predicted values are assigned to the corresponding query, according to which the correlation degree of the query is sorted, and the first k queries with the highest predictive value are recommended to the user. Secondly, based on the historical data of a single user, In addition, a recommendation technology of user query based on collaborative similar computing is proposed, which gathers user data with the same search behavior to target user. This method can solve the problem that the data quantity of target user is insufficient. In addition, it can provide users with certain expansibility and novelty of query recommendation. In the similarity calculation of users, the history query log of each user is integrated into a document, and the similarity between documents is calculated by using cosine vector model. Then using the cooperative recommendation theory in the recommendation system, the user's click frequency ratio to each link is taken as the relevant preference score. Based on these click scores, the improved Euclidean distance is used to calculate the similarity between users. Finally, the two similar computing methods are integrated in a linear weighted way, and the similar user sets of target users are calculated. Finally, the click model is used for query recommendation. Finally, based on the analysis of query recommendation, A simple personalized search engine system is implemented. In this system, the related query recommendation algorithm is introduced into the system, and the user click model is applied to the web page sorting, and the personalized query recommendation and web page sorting functions are realized.
【学位授予单位】：江南大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】