Aggressive Morphology and Lexical Relations for Query Expansion
Aggressive Morphology and Lexical Relations for Query Expansion
复制标题
用于查询扩展的积极形态学和词汇关系
DOI：
--
复制
发表时间：
2001
期刊：
影响因子：
--
通讯作者：
A. Houston
中科院分区：
文献类型：
--
作者：
W. Woods;Stephen Joseph Green;P. Martin;A. Houston
我们今年向TREC提交的申请是基于一系列系统的组合。第一个是概念索引和检索系统，它是由Sun Microsystems实验室开发的（Woods等人，2000 a; Woods等人，2000 b）。第二个是滑铁卢大学开发的多文本系统（Clarke等人，2000; Cormack等人，2000年）。概念索引系统旨在帮助人们在不受限制的文本中找到特定问题的特定答案。它结合了句法、语义和形态学知识，以及分类学的包容技术，来解决用户查询和可能回答它们的材料之间的术语差异。在索引时，系统建立索引材料中所有单词和短语的概念分类。这种分类是基于词的形态结构，短语的句法结构，以及它在词典中所知道的词的含义之间的语义关系。然而，它并不是作为一个问答系统设计的。我们去年的结果虽然令人鼓舞，但表明我们需要在问题分析方面做更多的工作（即，“这个问题的答案是什么？”）以及答案确定（即，“这段检索到的文章实际上回答了这个问题吗？”）来支持我们的松弛排序段落检索算法在与滑铁卢大学的研究人员交谈后，我们决定提交一个运行，我们将提供前端处理，包括使用我们自动派生的分类法进行查询制定和查询扩展，滑铁卢将通过他们的多文本段落检索系统和答案选择组件提供后端处理。其结果是一个直接的比较两个问答系统，不同的只是在查询制定组件。
Our submission to TREC this year is based on a combination of systems. The first is the conceptual indexing and retrieval system that was developed at Sun Microsystems Laboratories (Woods et al., 2000a; Woods et al., 2000b). The second is the MultiText system developed at the University of Waterloo (Clarke et al., 2000; Cormack et al., 2000). The conceptual indexing system was designed to help people find specific answers to specific questions in unrestricted text. It uses a combination of syntactic, semantic, and morphological knowledge, together with taxonomic subsumption techniques, to address differences in terminology between a user’s queries and the material that may answer them. At indexing time, the system builds a conceptual taxonomy of all the words and phrases in the indexed material. This taxonomy is based on the morphological structure of words, the syntactic structure of phrases, and semantic relations between meanings of words that it knows in its lexicon. It was not, however, designed as a question answering system. Our results from last year, while encouraging, showed that we needed more work in the area of question analysis (i.e., “What would constitute an answer to this question?”) and answer determination (i.e., “Does this retrieved passage actually answer the question?”) to support our relaxation ranking passage retrieval algorithm. After conversations with the researchers at the University of Waterloo, we decided to submit a run where we would provide front-end processing consisting of query formulation and query expansion using our automatically derived taxonomy and Waterloo would provide the back-end processing via their MultiText passage retrieval system and their answer selection component. The result is a direct comparison of two question answering systems that differ only in the query formulation component.