基于信息抽取的食品安全事件自动问答系统方法研究

QA System for Food Safety Events Based on Information Extraction

针对从海量食品安全事件新闻报道中很难抽取出所需答案的问题,以食品安全事件语料库为研究对象,提出了一种基于信息抽取技术的自动问答系统。首先,利用深度学习模型TextCNN对用户输入的问题进行分类,得到其所属类型。其次,对于输入问题,借助Lucene搜索引擎找到其最佳匹配文档。再次,根据输入问题的类型,从食品安全事件数据库(采用信息抽取技术自动提取的一个结构化数据库)中筛选出该文档所包含的答案候选句集合。最后,利用深度学习模型Bi-LSTM及基于答案候选句上下文的特征提取方法构建一个答案抽取模型,该模型能从给定的答案候选句集合中提取出最终答案。为检查基于食品安全事件数据库的答案候选句筛选方式及基于答案候选句上下文的特征提取方式对整个自动问答系统性能的影响,进行了多种比较实验,结果表明含有基于食品安全事件数据库的答案候选句筛选方式和基于答案候选句上下文的特征提取方式的问答系统表现最佳,其回答准确率达到44%。这相比于传统的问答系统,具有明显的优势。

To solve the problem of extracting the answer for a question from massive food safety incident news reports,a question answering(QA)system was proposed.Firstly,a deep learning method TextCNN was used to classify the question provided by users.Secondly,a search engine method Lucene was used to find the best matching report for the question.Thirdly,based on a food safety event database(a structured knowledge base which was automatically constructed by using the information extraction technology)and the type of the input question,a set of answer sentence candidates was selected from the best matching report.Finally,based on the deep learning model of bi-directional long short-term memory(Bi-LSTM)and a feature extraction method which can extract effective information from the contexts of the answer candidate sentences,an answer extraction model was constructed,which can automatically extract the final answer from the given set of answer candidate sentences.To evaluate the impact of the selection method of answer candidate sentences based on the food safety event database and the feature extraction method based on the contexts of answer candidate sentences,different experiments were conducted.The results showed that the QA system using the structured knowledge database and the context-based feature extraction achieved the best performance(44%in accuracy),which significantly outperformed over traditional QA systems.