基于词项图分析的查询分面挖掘方法

发布时间：2018-03-01 17:36

本文关键词： 查询分面用户意图频繁列表词项图知识库社会媒体社会计算　出处：《计算机学报》2017年03期 　论文类型：期刊论文

【摘要】：查询分面是用于描述查询某一方面内容的一组并列的词或词组.现有的查询分面挖掘方法主要通过模式挖掘搜索结果中包含的高频列表,并利用无监督或有监督的方法对高频列表进行聚类,最终得到查询分面.因为通常采用的搜索结果的数目有限,这种方法挖掘出的查询分面及其包含的分面项的覆盖率不高.针对这一问题,该文提出了一种基于从大规模网页中构建的词项图的查询分面挖掘方法.首先基于大规模网页数据集构建词项图,图中的节点代表词项,边代表两个词项的相似性.针对每个查询,从搜索结果中挖掘出初始分面,然后基于词项图对这些初始查询分面进行扩充,找到词项图中与初始分面类似的候选词,对候选词抽取多种特征,最后利用支持向量机对候选词进行分类,预测词项是否可为扩充词项,并将预测为正例的词项扩充到分面中.该扩充过程迭代多次直到无法找到更多分面项.实验表明该方法可有效提高查询分面的质量,尤其是能够显著改善分面项的覆盖率.
[Abstract]:Query partitioning is a set of paratactic words or phrases used to describe a certain aspect of a query. And using unsupervised or supervised methods to cluster high-frequency lists, and finally get the query facets, because the number of commonly used search results is limited, This method does not have a high coverage of the query partitioning and the items it contains. In view of this problem, In this paper, a query face mining method based on the term graph constructed from large scale web pages is proposed. Firstly, a word item graph is constructed based on a large scale web page data set, and the nodes in the graph represent the word item. The edges represent the similarity of the two terms. For each query, the initial face is mined from the search results, then the initial query faces are expanded based on the term graph, and the candidate words similar to the initial face are found in the term graph. Finally, support vector machine (SVM) is used to classify candidate words to predict whether they can be extended. The term items predicted as positive examples are extended to the facets. The extended process iterates many times until no more facets can be found. Experiments show that this method can effectively improve the quality of query partitioning, especially the coverage of partitioning items.
【作者单位】：中国人民大学信息学院;中国艺术科技研究所;中国人民大学大数据管理与分析方法研究北京市重点实验室;
【基金】：国家自然科学基金(61502501) 国家“九七三”重点基础研究发展规划项目基金(2014CB340403)资助~~
【分类号】：TP391.1

【相似文献】