基于用户兴趣模型的个性化搜索系统的研究

发布时间：2019-02-23 15:35

【摘要】：快速发展的互联网在给人们提供大量信息的同时,也不可避免地让用户陷入难以快速获取有价值信息的窘态。搜索引擎已经成为获取网络资源的一种主要方式,但是在实际应用中,即使是不同背景和需求的用户,在输入相同查询词的时候也会得到相同的结果,甚至是相同的网页排序,因此基于用户兴趣偏好的个性化搜索服务是搜索引擎今后发展的一个方向。本文通过对个性化搜索引擎工作原理和基本技术的研究和探讨,构建了一个简单的基于用户兴趣模型的个性化搜索原型系统。该系统首先获取用户浏览历史记录,并对其进行预处理、中文分词和基于TF-IDF算法的特征词词频统计,利用空间向量表示方法来选取权重最高的N个关键词建立用户的兴趣模型。然后,系统采用余弦公式计算最初的搜索结果与该模型的相似度,来对搜索结果进行过滤,去除用户不感兴趣的信息,从而挑选出与用户查询匹配度高的网页,保证了网页内容的相关度；并通过分析用户对搜索结果的点击行为,提取搜索结果与查询关键词的相关性,进而增加相关结果页面的权重,降低不相关页面的权重；整合以上两个阶段的相关度评分结果,根据相关度大小对搜索结果进行过滤和重新排序,实现了面向单用户的个性化搜索。最后,通过对用户浏览数据和点击行为的挖掘,借助于全文搜索引擎Lucene和开源搜索框架Solr,设计并实现了个性化搜索系统。实验过程中与未实现个性化搜索的系统作对比,实验结果证明：基于用户兴趣模型的个性化搜索会根据用户兴趣模型和用户点击行为对搜索结果进行过滤和重排序,搜索结果符合用户的兴趣偏好,有效提高搜索引擎的查准率和用户使用满意度。
[Abstract]:The rapid development of the Internet not only provides people with a lot of information, but also inevitably makes it difficult for users to obtain valuable information quickly. Search engine has become one of the main ways to obtain network resources, but in practical applications, even users with different backgrounds and needs will get the same results when entering the same query words, even the same ranking of web pages. Therefore, personalized search service based on user interest preference is a direction of search engine development in the future. Based on the research and discussion of the working principle and basic technology of personalized search engine, a simple personalized search prototype system based on user interest model is constructed in this paper. The system firstly acquires the user's browsing history record and preprocesses it. The Chinese word segmentation and the feature word frequency statistics based on TF-IDF algorithm. The interest model of users is established by using spatial vector representation method to select N keywords with the highest weight. Then, the system uses cosine formula to calculate the similarity between the initial search results and the model, to filter the search results, remove the information that users are not interested in, and then select the web pages with high matching degree with the users. Ensure the relevance of web content; By analyzing the user's click behavior on the search results, the relevance between the search results and the query keywords is extracted, and then the weight of the relevant result pages is increased, and the weight of the unrelated pages is reduced. Integrating the results of the above two stages, filtering and reordering the search results according to the degree of relevance, the personalized search for single user is realized. Finally, a personalized search system is designed and implemented with the help of full-text search engine (Lucene) and open source search framework (Solr,) by mining the user's browsing data and clicking behavior. The experiment results show that the personalized search based on user interest model will filter and reorder the search results according to user interest model and user click behavior. The search results accord with the user's interest preference, and improve the precision and user satisfaction of search engine effectively.
【学位授予单位】：西南大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.3

【参考文献】