arXiv:2004.03101 · 中英对照阅读
面向开放域问答的知识融合与语义知识排序
Knowledge Fusion and Semantic Knowledge Ranking for Open Domain Question Answering
中文速览
回答科学选择题时，系统既要从海量文本里找对事实，还得把分散的事实拼起来推理，常常会被检索到的无关信息干扰。研究者先用搜索引擎分两步找知识，再训练基于 BERT 的语义排序模型筛掉不相关事实，并设计知识融合模块，让模型能结合外部事实、比较不同答案选项来作答。在 OpenBookQA 和 QASC 两个数据集上，这套方法都超过了此前结果，准确率分别提升 2.2 和 7.28 个百分点。它说明，光有强大的语言模型还不够；更有针对性地挑选、融合外部知识，能让多步推理型问答更可靠。
摘要
开放域问答要求系统检索外部知识，并通过组合分散在多个句子中的知识进行多跳推理。在近期推出的开放域问答挑战数据集中，包括 QASC 和 OpenBookQA，我们需要检索事实并组合事实，以正确回答问题。在我们的工作中，我们训练了一个语义知识排序模型，对通过基于 Lucene 的信息检索系统检索到的知识重新排序。我们进一步提出了一种“知识融合模型”，该模型利用基于 BERT 的语言模型中的知识与外部检索到的知识，从而提升基于 BERT 的语言模型对知识的理解。在 OpenBookQA 和 QASC 两个数据集上，使用经过语义重排序的知识的知识融合模型均优于此前的方法。
Open Domain Question Answering requires systems to retrieve external knowledge and perform multi-hop reasoning by composing knowledge spread over multiple sentences. In the recently introduced open domain question answering challenge datasets, QASC and OpenBookQA, we need to perform retrieval of facts and compose facts to correctly answer questions. In our work, we learn a semantic knowledge ranking model to re-rank knowledge retrieved through Lucene based information retrieval systems. We further propose a “knowledge fusion model” which leverages knowledge in BERT-based language models with externally retrieved knowledge and improves the knowledge understanding of the BERT-based language models. On both OpenBookQA and QASC datasets, the knowledge fusion model with semantically re-ranked knowledge outperforms previous attempts.
术语表
- Open Domain Question Answering
- 开放域问答
- Natural Language Question Answering
- 自然语言问答
- OpenBookQA
- OpenBookQA
- QASC
- QASC
- HotPotQA
- HotPotQA
- Natural Questions
- Natural Questions
- MultiRC
- MultiRC
- ComplexWebQuestions
- ComplexWebQuestions
- WikiHop
- WikiHop
- SciTail
- SciTail
- multi-hop reasoning
- 多跳推理
- knowledge retrieval
- 知识检索
- knowledge composition
- 知识组合
- knowledge corpus
- 知识语料库
- semantic knowledge ranking
- 语义知识排序
- knowledge fusion model
- 知识融合模型
- BERT
- BERT
- Lucene
- Lucene
- Elasticsearch
- Elasticsearch
- information retrieval
- 信息检索
- ranking model
- 排序模型
- multiple-choice question answering
- 多项选择问答
- supervised knowledge retrieval model
- 有监督知识检索模型
- external knowledge
- 外部知识
- knowledge-enriched embeddings
- 知识增强嵌入
- neural explanation retrieval
- 神经解释检索
- state-of-the-art
- 当前最优