Complex Interactive Question Answering Enhanced with Wikipedia

TREC(2007)

引用 24|浏览31
暂无评分
摘要
In ciQA, templates are used with several bracketed items we call ”facets” which are the basis of information being sought. This information is returned in the form of nuggets. Due to the concepts being sought having multiple terms to describe them, it becomes difficult to determine which sentences in the AQUAINT-2 news articles contain the query terms being sought, as they may be represented in the parent document by a variety of different phrases still making reference to the query term. For example, if the term ”John McCain” were being sought, it might appear in an article, however, the sentence which is the vital nugget may simply contain ”Senator McCain”; an imperfect match. Traditional query expansion[5] of facets would introduce new terms which are related but do not necessarily mean the same as the original facet. This does not always help the problem of query terms appearing in relevant documents but not relevant sentences within documents, it only introduces related terms which cannot be considered synonymous with the facet being retrieved. In this year’s TREC, we hope to overcome some of this problem by looking for synonyms for facets using Wikipedia. Many of the ciQA facets are proper nouns and most thesauri, such as WordNet, do not contain entries for these. Thus, a new manner of finding synonyms must be found. In recent years, several new approaches have been proposed to use Wikipedia as a source of lexical information[2, 7], as it can be downloaded in its entirety, and contains relatively high quality articles[3]. 2 Wikipedia
更多
查看译文
关键词
Information Retrieval,Topic Modeling,Named Entity Recognition
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要