摘要
单文档问答式信息检索,即是阅读理解(Reading Comprehension,简称RC)。该任务的目的在于理解一篇文档并对提出的问题返回答案句。提出了充分利用外部资源采用多策略技术来提高RC系统性能的方法,包括基于Web的答案模式匹配应用、词汇语义关联推理以及上下文辅助等策略。本方法使得RC系统性能在Remedia标准测试集上的性能得到提高。描述了不同策略对提高系统性能的有效性,t-test结果表明,运用答案模式匹配和词汇语义关联推理策略所得到的性能显著提高;同时分析了指代消解策略在系统中的关键作用;最后比较了RC任务和多文档问答式信息检索(Question Answering,简称QA)任务的差异性。
Single document question answering is also called Reading Comprehension（RC）, which attempts to understand a document and returns an answer sentence when posed with a question. We proposed an approach that adopted multi-strategy and utilized external knowledge to improve the performance of RC, including pattern matching with Web- based answer patterns, lexical semantic relation inference and context assistance. This approach gives improved RC performance on the Remedia corpus. The effectiveness of different strategy was analyzed and pairwise t-tests show the performance improvements due to Web-derived answer patterns and lexical semantic relation inference technique are statistically significant. In addition, the performance impact by the co-reference resolution was also discussed. Finally, the comparison between the task of RC and multi-document question answering（QA） was analyzed.
出处
《计算机科学》
CSCD
北大核心
2009年第7期193-196,共4页
Computer Science
基金
国家自然科学基金青年基金(No.60803086)
北京工业大学博士科研启动基金(52007012200701)资助
关键词
模式
阅读理解
问题回答
自然语言处理
Pattern, Reading comprehension, Question answering, Natural language processing
