Joint Recognition Of Bus Travel Intention And Semantic Slot Filling Based On Attention+BLSTM

With the development of computing technology and artificial intelligence technology,the oral dialogue system has been more and more widely used,especially the oral dialogue system for specific tasks,not only can provide people with convenient consulting services anytime,anywhere,but also for people's lives bring a lot of convenience.As an important part of the dialogue system,oral comprehension usually involves two tasks: intent recognition and semantic slot filling.The traditional speaking understanding module usually processes two tasks separately.Intent recognition usually uses a classification method to classify sentences into corresponding intent categories.Semantic slot filling is to sequentially label each word in a given sentence.At present,the joint recognition of intent and semantic slot filling has become the mainstream method of oral language understanding research.Based on this,this article uses the path-finding sentence as the experimental data for the public transport field in Hohhot,conduct oral comprehension studies,identify the starting point and destination in the path-finding sentence as a semantic slot filling,and judge the travel intent of the entire sentence to understand the user's travel needs.The details are as follows:(1)Aiming at the problem that the traditional Conditional Random Field(CRF)model cannot set the intent label,a method of converting intent recognition into a sequence labeling task is adopted.By adding a flag bit to the word or word in the sentence to represent the intent,in order to realize the joint recognition of intent and semantic slot filling using the CRF model.(2)Aiming at the problem that the existing intent and semantic slot filling joint recognition method does not extract semantic information well,adopting a Bidirectional long short term memory(BLSTM)model based on attention mechanism not only avoids the problem of gradient disappearance of RNN,but also based on LSTM It can also obtain backward semantic information.After adding the attention mechanism,the required semantic information can be obtained more accurately.The experimental data in this paper uses the bus route query sentence in Hohhot as the corpus,identifies the origin and destination in the route query sentence as the semantic slot filling,judges the travel intention of the sentence as a whole,and understands the user's travel needs.Experimental results show that the BLSTM model based on the Attention mechanism is superior to the three methods of LSTM,BLSTM,LSTM + Attention in the accuracy of intent recognition and the F1 value of semantic slot filling,and in each method,the results based on character tags are better than the results based on word tags.Compared with the CRF joint recognition model,the F1 value of the semantic slot filling is higher than that of the CRF method,and the accuracy of intent recognition is lower than that of the CRF method.Overall,the BLSTM model based on the Attention mechanism achieved the best results.