基于机器阅读理解的生活情景常识预测

General Knowledge Prediction of Life Situations Based on Machine Reading Comprehension

摘要: 机器学习研究的长期目标是产生适用于推理和自然语言的方法，建立智能对话系统。本实验通过回答日常生活的事件的问答问题来评估阅读理解，使用Facebook AI的BABI tasks中的四种类型数据完成模型训练，采用数字编码稀疏交叉熵损失函数对RNN模型、LSTM模型和BERT模型参数进行设置，采用多分类单标签的categorical_accuracy函数作为评价度量，预测样本数据集中的正确数量。实验结果表明，在RNN模型预测答案的准确率明显高于LSTM和BERT模型。

Abstract: The long-term goal of machine learning research is to generate methods applicable to reasoning and natural language to build intelligent conversational systems. This experiment evaluates reading comprehension by answering question-and-answer questions about everyday events. Model training is completed using four types of data from Facebook AI’s BABI tasks, and the RNN model, LSTM model, and BERT model parameters are set using a digitally encoded sparse cross-entropy loss function, and a multicategorical single-label categorical_ accuracy function is used as an evaluation metric to predict the number of corrections in the sample dataset. The experimental results show that the accuracy of predicting answers in the RNN model is significantly higher than that of the LSTM and BERT models.