基于搜索日志的用户语义本体构建研究

发布时间：2018-02-11 08:43

本文关键词： 用户语义本体用户日志概念格形式概念分析 WordNet　出处：《西华大学》2012年硕士论文　论文类型：学位论文

【摘要】：近年来，随着互联网信息技术的迅速发展，互联网上的信息资源已经达到海量级别，并且正呈指数级形式增长,海量网页数据的结构复杂性和用户查询词简短及语义模糊性等特点给现有搜索引擎的发展带来了极大的挑战。检索系统如何能够准确的理解用户输入查询词的信息需求，根据不同用户返回不同的检索结果，即根据不同的用户提供个性化的服务，这是用户越来越关心的问题。要为用户提供个性化的服务，就必须挖掘用户的相关领域知识背景，为搜索引擎提供一个面向用户的知识模型，即用户本体。一般情况下，搜索引擎搜集了大量的用户搜索日志，这些数据记录了用户历史查询词和对应点击网页的信息，通过分析这些历史数据，能够挖掘出用户的领域背景知识。本体（Ontology）作为语义网的关键技术，它通过提供一个领域的词汇和形式概念，使得信息的共享和交互变得容易和简单。本文的主要工作如下：首先，本文提出了一种新颖的用户查询词语义相似度计算方法，采用AGNES(Agglomerative Nesting)层次聚类算法，将用户查询词按用户个性化兴趣和知识背景进行主题分类。本文首先提出了基于用户搜索日志的三种用户查询词语义相似关系①基于用户原始查询词本身的相似关系，②基于用户扩展查询词的相似关系，③基于用户点击URLs的相似关系。通过分析这三种语义关系，将它们按照线性组合的方式组合，形成了一种新颖的计算用户查询词语义相似度的方法。基于这种用户查询词语义相似度函数，利用AGNES层次聚类算法，将用户查询词根据用户搜索日志中所反映的主题进行语义主题聚类，从而达到消除用户查询词语义模糊性的目的。其次，本文提出了一种利用用户查询词语义主题聚类结果和WordNet通用本体建立一个用户查询词兴趣主题领域知识模型，即用户语义本体（User Semantic Ontology）的方法。该方法过程如下①根据用户查询词语义主题聚类结果，生成用户原始查询词-用户点击文档和扩展查询词-用户点击文档之间的形式背景，②优化扩展查询词-用户点击文档之间的形式背景，合并原始查询词和优化后的扩展查询词的形式背景，并构建概念格，通过概念格-本体转化的规则，将概念格转换为初始本体，③利用WordNet优化初始本体。该用户本体表达了一个用户的兴趣偏好，然后将其应用于主题搜索引擎，，进而可以把信息采集从基于关键词的相关度匹配技术层面提高到基于语义层面的查找。最后，利用VC++6.0开发应用程序进行验证。实验表明，通过本文本体构建方法，用户查询词能更好的根据用户兴趣和知识背景来区分其真实语义，消除其语义模糊性。
[Abstract]:In recent years, with the rapid development of Internet information technology, the information resources on the Internet have reached a mass level, and are growing exponentially. The structural complexity of massive web page data and the short and semantic fuzziness of user query words bring great challenges to the development of existing search engines. How can the retrieval system accurately understand the information requirements of user input query words? According to different users return different retrieval results, that is, according to different users to provide personalized services, which is increasingly concerned by users. In order to provide personalized services for users, we must dig up the relevant domain knowledge background of users. In general, the search engine collects a large number of user search logs, which record the user history query words and the corresponding information of clicking on the web page, which provides a user-oriented knowledge model for the search engine. By analyzing these historical data, we can mine the domain background knowledge of the user. Ontology (Ontology) as the key technology of semantic Web, it makes the sharing and interaction of information easy and simple by providing the vocabulary and formal concept of a domain. The main work of this paper is as follows:. First of all, this paper proposes a novel method to calculate the semantic similarity of user query words, using AGNES(Agglomerative clustering hierarchical clustering algorithm. The user query words are classified according to the user's personalized interest and knowledge background. Firstly, three kinds of semantic similarity relation of user query words based on user search log are proposed. 1 based on the similarity of the original query words of the user, this paper proposes three kinds of semantic similarity relation of user query words based on user search log. Relationship 2 is based on the similarity relation of user extended query words. 3 based on the similarity relation of user clicking on URLs, this paper analyzes the three semantic relationships. A novel method of calculating the semantic similarity of user query words is formed by combining them according to linear combination. Based on this function of semantic similarity, AGNES hierarchical clustering algorithm is used. The user query words are clustered according to the topics reflected in the user search log, so as to eliminate the ambiguity of the meaning of the user query words. Secondly, this paper proposes a knowledge model of topic domain of user query word interest, which is based on the result of semantic clustering of user query words and WordNet general ontology. The process of this method is as follows: (1) clustering result according to the user query word meaning topic, Generate user original query words-user click document and extended query word-user click formal background between documents / optimize extended query word-user clicks on formal background between documents, The formal background of the original query words and the optimized extended query words is combined, and the concept lattice is constructed, which is transformed by the rules of concept lattice-ontology transformation. The concept lattice is transformed into the initial ontology 3, which uses WordNet to optimize the initial ontology. The user ontology expresses a user's interest preference and then applies it to the subject search engine. Furthermore, the information collection can be improved from keyword based correlation matching technology to semantic level based search. Finally, using VC 6.0 to develop the application program to verify. The experiment shows that the user query words can better distinguish their real semantics and eliminate their semantic fuzziness according to the user's interest and knowledge background through the method of ontology construction in this paper.
【学位授予单位】：西华大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.1

【相似文献】