一种建立在对客户端浏览历史进行LDA建模基础上的个性化查询推荐算法

发布时间：2018-06-10 16:45

本文选题：查询推荐 + 个性化　；参考：《四川大学学报(自然科学版)》2015年04期

【摘要】：现代搜索引擎普遍采用简单的关键词形式来表达查询,这为用户带来便利的同时也增加了准确获取信息的难度.搜索引擎很难基于少量的几个关键词准确捕捉用户的信息需求.查询推荐作为缓解上述问题的关键技术,已经开始应用于目前主流的搜索引擎.然而,绝大多数现有的查询推荐技术基于群体智慧,以搜索引擎日志为数据源,从中挖掘用户群体在构造查询方面的习性及查询之间的语义关联性,未考虑不同用户个性化的信息喜好,而且在搜索引擎服务器端进行查询推荐计算也会影响搜索引擎的响应效率和查询吞吐率.为此,本文提出一种运行于客户端的个性化查询推荐策略,该策略以用户浏览历史为数据源,采用LDA模型,并从中学习用户的信息喜好,在此基础上利用原始查询被主题模型生成的概率确定用户的查询意图,以词条与查询意图之间的关联来度量词条作为扩展查询被推荐的强度,最后选择Top N强度最强的词条作为查询推荐的结果.以人工标注的测试数据对本文所提出的算法进行了实验验证,结果表明,该算法在推荐扩展查询的准确率方面明显优于单纯基于词条与原始查询语义相关的方法.
[Abstract]:Modern search engines generally use simple keywords to express queries, which brings convenience to users, but also increases the difficulty of accurately obtaining information. It is difficult for search engines to accurately capture the information needs of users based on a few keywords. Query recommendation, as a key technology to alleviate the above problems, has been applied to the current mainstream search engine. However, most of the existing query recommendation techniques are based on group intelligence and take search engine logs as data sources to mine the habits of user groups in constructing queries and the semantic correlations between them. The information preferences of different users are not taken into account and the search engine response efficiency and query throughput will be affected by the query recommendation calculation on the server side of the search engine. Therefore, this paper proposes a personalized query recommendation strategy running on the client side. The strategy takes the user's browsing history as the data source, adopts the LDA model, and learns the information preference of the user. On this basis, the user's query intention is determined by the probability that the original query is generated by the topic model, and the relationship between the entry and the query intention is used to measure the strength of the extended query as the recommended term. Finally, the top N strongest term is selected as the result of query recommendation. The experimental results show that the proposed algorithm is superior to the original query based on terms and expressions in the accuracy of recommendation extended query.
【作者单位】：四川大学计算机学院;浙江师范大学数理与信息工程学院;中国民用航空总局第二研究所信息技术分公司;
【基金】：浙江省自然科学基金(LY12F02010) 四川省科技支撑项目(2014GZ0063)
【分类号】：TP391.3

【相似文献】