首页> 外文期刊>ACM SIGIR FORUM >Passage Retrieval vs. Document Retrieval for Factoid Question Answering
【24h】

Passage Retrieval vs. Document Retrieval for Factoid Question Answering

机译:Factoid问题解答的段落检索与文档检索

获取原文
获取原文并翻译 | 示例

摘要

QA systems are usually evaluated end-to-end, on the basis of their ability to find answers to questions. Unfortunately, QA systems are complex ― with many subsystems ― and an end-to-end evaluation often does not allow us to understand the relative contribution that each subsystem makes to the overall performance. Here, we examine the IR subsystem in isolation. Our results suggest that full-document retrieval may be desirable in a QA system, since higher precision may translate into better performance. However, when full documents are retrieved, post-retrieval passage selection may still be required. The MultiText method, even when used for full-document retrieval, identifies a hotspot where the answer is more likely to found, combining the advantages of both passage and document retrieval.
机译:质量检查系统通常根据其找到问题答案的能力进行端到端评估。不幸的是,质量保证系统很复杂-包含许多子系统-并且端到端评估通常不能使我们了解每个子系统对整体性能的相对贡献。在这里,我们单独检查IR子系统。我们的结果表明,由于更高的精度可能会转化为更好的性能,因此在QA系统中可能需要使用全文档检索。但是,在检索完整文档时,仍可能需要选择检索后的段落。 MultiText方法,即使用于全文档检索,也结合了段落和文档检索的优点,可以识别出更容易找到答案的热点。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号