Abstract: Pseudo relevance feedback (PRF) has been perceived as an effective solution for automatic query expansion. However, traditional pseudo relevance feedback can result in the query representation “drifting” away from the original query and a decreased retrieval performance. Therefore, the key issues in applying PRF are to identify the real relevant documents in the top retrieved results without any other assistant information, and expend the query based on the these relevant documents. This paper presents a solution framework from extensible markup language (XML) data. Firstly, this paper considers the XML content and structure features, and proposes a good XML query scheme based on pseudo relevance feedback documents by combining search results clustering with a two-stage ranking model. Furthermore, this paper explores the XML query expansion of CO (content only) query, and gives the term weight computation with structure. The experimental results show that the proposed scheme can reduce the topic drift effectively and obtain the better retrieval quality.

Key words: XML pseudo relevance feedback, search results clustering, ranking, query expansion

摘要： 伪反馈（pseudo relevance feedback，PRF）一直以来都被认为是一种有效的查询扩展技术。然而传统的伪反馈容易带来主题漂移，从而影响检索性能。如何确定高质量的相关文档集，以及如何从相关文档集中挑选有用的扩展词项，是解决伪反馈中查询主题漂移的两个重要方面。对此，针对XML（extensible markup language）文档，提出了一个解决框架：一方面，研究了XML伪反馈文档查找方法，在充分考虑XML内容和结构特征的前提下，提出了基于检索结果聚类和两阶段排序模型相结合的高质量XML伪相关文档查找技术；另一方面，针对CO（content only）查询，对词项扩展进行了研究，提出了带结构语义的词项权值计算方法。一系列的相关实验数据表明，所提的XML伪反馈查询扩展方法能有效地减少查询主题漂移现象，获得更好的检索质量。

关键词: XML伪反馈, 检索结果聚类, 排序, 查询扩展