基于虚拟技术和深度学习英语教学训练系统设计

Based on Virtual technology and Deep Learning

针对传统英语教学训练系统对于英语语义理解准确率差,造成英语口语对话和教学训练效果不佳的问题,设计一个基于虚拟技术和深度学习的英语教学训练系统。在BERT方法和BiLSTM神经网络的基础上,构建一个基于角色信息的语义理解模型。此模型采用BERT方法对输入语句进行预处理后,提取语句中的嵌入词向量,再通过多层感知机计算参与编码的历史语句权重值。最后利用BiLSTM网络提取历史语句信息,并通过Softmax函数获取语句意图概率信息,由此实现基于意图识别的口语对话。最后将此模型应用到构建的英语口语训练的对话系统中,通过虚拟角色对话方式实现口语练习。实验结果表明,相较于传统的口语理解方法,本方法的意图识别率和槽值填充分别取值为75.35%和70.08%,均高于其他模型。且系统应用发现,通过虚拟技术可为学生提供虚拟人物进行对话和互动,提高了英语口语练习和对话,具有一定的有效性。

The traditional English teaching and training system is poor in the accuracy of English semantic understanding, causing poor oral English dialogue and teaching and training effect, and an English teaching and training system based on virtual technology and deep learning is designed. Based on the BERT method and the BiLSTM neural network, we construct a semantic understanding model based on role information. This model uses the BERT method to preprocess the input statement, and the embedded word vectors in the statement are extracted, and then the weight value of the historical statement involved in the encoding is calculated through the multi-layer perceptron. Finally, the BiLSTM network is used to extract the historical statement information, and the statement intention probability information is obtained through the Softmax function, thus realizing the verbal dialogue based on intention recognition. Finally, this model is applied to the constructed dialogue system of oral English training to realize oral practice through virtual character dialogue. The experimental results show that the intention recognition rate and Cao Zhi filling are 75.35% and 70.08%, respectively, which are higher than the other models. And the system application found that through the virtual technology can provide students with virtual characters to have a dialogue and interaction, improve the oral English practice and dialogue, has a certain effectiveness.