Research on Automatic Question Answering System for Online Educational Robots Based on Deep Learning

A machine reading comprehension model based on MacBERT and BiLSTM is proposed to improve the accuracy of the automatic question-answering system of online educational robots. On the one hand, the model embeds and encodes the input text through the dynamic word vector of MacBERT. On the other hand, it uses BiLSTM with contextual information learning ability to extract features from vector samples, thereby further improving the generalization performance of the model. The results reveal that the proposed model has higher semantic similarity matching ability for different types of questions. Compared with the traditional language models such as Word2vec and Seq2seq, the F1 value of the proposed model is increased by 4.01% and 3.25%, and the response accuracy is higher. The system test results show that the average response time and maximum response time of the automatic question answering system for online educational robots based on MacBERT are only 0.35s and 0.71s in the responses to 120 questions, and the response speed is fast. In addition, the accuracy of the response reaches 89.17%, and the most of the generated answers meet users' learning requirements, which is worth further research and promotion.