Design and Implementation of an Automatic Question and Answer System Based on BERT Model
To address the situation where problems encountered during online course learning cannot be answered in a timely manner,an automatic question and answer system is designed and implemented.Firstly,real problems in the course are collected as training datasets.Then,a two-tower neural network model is constructed based on the BERT model.It enters the problem in pairs into the model,and the training objective is to train the feature vectors of semantically similar questions to be as similar as possible.After training the parameters of model,the values on accuracy and F1-Score performance indicators reaches 0.931 and 0.918,respectively.This paper uses the trained model to convert the problem sets and the questions proposed by learners into feature vectors,and uses Faiss to recall the most similar questions to the learner's questions in the feature vector sets.Finally,it returns the corresponding answers of the most similar questions.The system has high accuracy and effectiveness,which can provide support for online course learning.
automatic question and answer systemBERT modelsemantic similarityonline learning
万方数据
维普
