基于Seq2Seq的生成式自动问答系统应用与研究

Application and Research on Generative Automatic Question Answering System Based on Seq2Seq

由于要处理复杂的推理问题,同时要求处理输入序列生成输出序列,生成式自动问答系统被视为成为自然语言处理中一项困难的任务。Seq2Seq模型的提出为序列到序列的生成模型提供一个新的结构基础。基于Seq2Seq基础结构设计并实现一个轻量级的生成式自动问答模型。通过一个综合检验自动问答系统性能的数据集b Ab I-10k,验证与分析基于Seq2Seq的生成式自动问答系统模型在20种不同类型的推理任务上性能。

Requiring complex reasoning abilities, processing input sequence and generating output sequence, generative automatic question answer- ing system is regarded as a hard task in natural language processing. Seq2Seq provides a novel basic structure for the sequence-to-se- quence generation models. Designs and implements a generative automatic question and answer model which is based on Seq2Seq struc- ture. Evaluates on a synthesized question answering dataset bAbi-10k, analyzes the reasoning abilities of proposed question answering sys- tem on 20 different reasoning tasks.