面向检索式问答系统的阅读理解模型
首发时间:2018-12-07
摘要:在面向检索式问答系统的问答任务中,信息检索方法对问答任务的处理依旧停留在浅层语义,并且存在着优化难度大和优化成本高等问题。端到端方法具有结构简单和效果直观等优点,其也被越来越多地应用于问答系统优化。本文将阅读理解模型应用于面向检索式问答系统的问答任务中,并针对该任务的特点设计了一个面向检索式问答系统的阅读理解模型(Reading Comprehension model for Retrieval-based Question Answering, RQA-RC)。在面向检索式问答系统的问答任务中,针对目前阅读理解模型计算效率低的问题,本文设计了基于双向长短期记忆网络的问题编码结构和基于卷积神经网络的文档编码结构;针对该任务中文档存在的答案数量不固定的问题,本文设计了基于序列标注的预测结构。实验结果表明,RQA-RC模型在该任务上相比目前阅读理解模型在计算效率和预测效果上均有明显提升。
For information in English, please click here
Reading Comprehension model for Retrieval-based Question Answering system
Abstract:ForRetrieval-based Question Answering(Q&A) system, Information Research solution is still in the stage of dealing with simple semantic Q&A tasks. And there are problems such as high optimization difficulty and high optimization cost. The end-to-end method has the advantage of simple structure and intuitive effect which is also increasingly used in Q&A system optimization. This paper designs a Reading Comprehension model for Retrieval-based Q&A system. For Retrieval-based Q&A system, this paper designs a question encode structure based on Bi-directional Long Short-Term Memory and a document encode structure based on Convolutional Neural Network for the low computational efficiency of Reading Comprehension models. This paper designs a prediction structure based on Sequence Labeling for the characteristics of the number of answers in a document that are not fixed. The results of experiment show that our model has a significant improvement in computational efficiency and prediction effect compared to current Reading Comprehension models.
Keywords: Artificial Intelligence Question Answering system Reading Comprehension Sequence Labeling
基金:
引用
No.****
同行评议
勘误表
面向检索式问答系统的阅读理解模型
评论
全部评论0/1000