Research On Natural Language Understanding Of Air Travel Based On Joint Modeling

In recent years,with the rapid development of artificial intelligence technology,various dialogue systems have been widely studied and applied.Especially in specific areas of dialogue system,It can be applied to book ticket and query flight in the field of air travel.It is of great research and application value to reduce the burden of enterprises and operating costs.Natural language understanding is the core module of the dialogue system.Its purpose is to extract the user's intention and the important semantic information related to the intention from the sentence,and express it into a structured form that the computer can understand.In a specific domain,natural language understanding can be divided into two subtasks,intention recognition and semantic slot filling.The main research contents of this paper are as follows:Firstly,In this paper,This paper proposes a dual decoder model based on the attention mechanism to jointly recognize two subtasks of natural language understanding.Intent recognition and semantic slot filling are two subtasks corresponding to one decoder respectively.During training,the loss values of the two decoders will be transmitted to the encoder at the same time.Through continuous iterative training,the two tasks are finally optimized.Attention mechanism is added to further improve the effect of the model.The experimental results show that effect of the joint model proposed in this paper is far better than that of the single model,and compared with several previous joint models,the effect of intention recognition and semantic slot filling is better.Secondly,This paper proposes a text augmentation method based on word fine granularity.We use a certain mechanism to randomly remove the words with less information in the sentence.Then we use the words with similar semantics to replace the words in the sentence with a certain probability.This method not only increases the amount of data,but also increases the diversity of sentences,effectively alleviates the problem of over fitting caused by sparse samples,and improves the anti-interference ability of the model.Experiments show that the text augmentation method can improve the effect of intention recognition and semantic slot filling.Thirdly,This paper applies the BERT model to the task of natural language understanding in the field of air trave.Experiments show that the BERT model can improve the effect of the model in the task of natural language understanding.Finally,the research method is applied to the natural language understanding system in the field of air travel.