网络日志挖掘及其在查询理解中的应用研究

发布时间：2019-06-14 18:41

【摘要】：查询推荐系统是现代搜索引擎系统中的一项非常重要的技术,该技术推荐更符合用户需求的查询词以提高用户的搜索体验,目前许多方法都是通过计算用户查询词之间的相似度来对查询词进行聚类,然后对簇中的查询词按照某种策略推荐,其中没有考虑到查询词之间所隐含的属性信息,使得查询词的相似度计算方法中缺少部分辅助信息,同时推荐过程中仅考虑了用户的历史点击行为。为解决上述问题,本文采用网络查询日志和百科作为构建属性知识库的数据,并构建基于属性知识库的查询推荐系统。该方法通过抽取查询日志中的人名实体及百科中实体属性知识,形成了包含属性知识的人名知识库,根据高质量的属性规则和统计分类方法对知识库中的实体进行分类,基于不同的人物分类知识库,对人名实体进行有效推荐。本文主要创新点如下：依据建立的人物属性知识库,制定高质量的人物分类规则,对未标记的人物实体进行分类,其中对一词对类的实体,本文采用基于搜索引擎返回结果来对实体进行分类标记。提出一种基于信息重合度的人名实体分类算法,此方法为概率统计分类算法,是对基于规则的分类方法的补充,可以有效解决基于规则分类算法召回率低的缺陷。提出一种基于簇的用户兴趣模型,通过训练数据,使其具有查询推荐功能,可以根据用户兴趣将查询推荐给用户。实验结果表明：基于规则加统计的分类算法的准确率和召回率要高于只基于规则或统计的分类方法。与传统的查询推荐系统相比,基于簇的用户兴趣模型推荐方法可以很好的提高准确率,更好地反应了用户的真实查询意图。
[Abstract]:Query recommendation system is a very important technology in modern search engine system. This technology recommends query words that are more in line with the needs of users in order to improve the search experience of users. At present, many methods cluster query words by calculating the similarity between user query words, and then recommend the query words in the cluster according to a certain strategy, in which the attribute information implied between query words is not taken into account. Because of the lack of some auxiliary information in the similarity calculation method of query words, only the historical click behavior of users is considered in the recommendation process. In order to solve the above problems, this paper uses network query log and encyclopedia as the data to construct attribute knowledge base, and constructs a query recommendation system based on attribute knowledge base. By extracting the attribute knowledge of human name entities in query log and entity attribute knowledge in encyclopedia, the name knowledge base containing attribute knowledge is formed. According to the high quality attribute rules and statistical classification methods, the entities in the knowledge base are classified, and the named entities are effectively recommended based on different character classification knowledge bases. The main innovations of this paper are as follows: according to the established character attribute knowledge base, the high quality character classification rules are formulated, and the unmarked character entities are classified, in which the entity is classified and marked based on the results returned by the search engine. A classification algorithm of human name entities based on the degree of information coincidence is proposed. This method is a probability and statistical classification algorithm, which is a supplement to the rule-based classification method and can effectively solve the defect of low recall rate of the rule-based classification algorithm. In this paper, a cluster-based user interest model is proposed. By training the data, it has the function of query recommendation, and the query can be recommended to the user according to the user's interest. The experimental results show that the accuracy and recall rate of the classification algorithm based on rule plus statistics are higher than those based only on rules or statistics. Compared with the traditional query recommendation system, the cluster-based user interest model recommendation method can improve the accuracy and better reflect the real query intention of the user.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.3

【相似文献】