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.