Research On Indentification Method Of The User's Travel Consumption Intention In Chatting Robot

With the rapid development of computer technology,especially artificial intelligence technology,chatting robot,a new form of intelligent dialogue system,has emerged and popularized.Based on this,the task of identifying the consumer's travel intention put forward in this paper.The so-called travel intention,refers to the chatting robot,users in order to meet the needs of travel,through the text to express the travel products or services purchase intention.Identifying the user's travel intention can not only enrich the conversation strategy of the chatting robot,but also help the product recommendation after the work.It is of great significance.We will travel consumption intention recognition task as a classification problem in this paper,the first use of feature engineering based machine learning method for travel intention recognition,through the analysis,the user's chat text is short and the spoken language,it is very difficult to identify.Therefore we attempts to use the co-occurrence relation mining Apriori algorithm and LDA algorithm for mining theme,chat text on the user to expand the two aspects of content and meaning of the topic,the expression of rich and perfect user chat information,make the model easier to recognize the user's intention to travel.The experimental results show that the recognition strategy based on content and topic expansion can improve the final result.Because machine learning based on feature engineering is very human and has strong limitations,it is still difficult to describe the deep semantic information of text even if it is extended.Then,we attempts to use end to end depth learning model for feature extraction and trip intention recognition.Specifically,we constructs the long-term memory neural network based on convolutional(Convolutional-LSTM)model of travel consumption intention feature mining and recognition,firstly by convolutional neural network(CNN)to extract the features of text chat users,then combined into long-short term memory neural network(LSTM)for feature representation learning,finally the output of the recognition results.The experimental results show that the travel consumption intention recognition task,the Convolutional-LSTM model compared with the traditional method based on the characteristics of engineering has greatly improved,and compared the performance of CNN and LSTM model has the advantage in the travel consumption intention recognition task is effective.There are many domains in travel intention.In the process of actual intention recognition,the corpus of some domain of intent is very scarce,and it is difficult to obtain.This brings a lot of inconvenience to the field of intention recognition.We attempts to use the underlying parameter sharing and multi task learning two different transfer learning methods,using small scale annotation data in the new field of transfer learning intention,and achieved ideal results.Finally,we will study the content used in the chatting robot "Benben" in this paper,through the identification of the user chat text and interaction with the user,the user determines the travel intention.On the meridian test,the travel consumption intention module performs well in the chat ting robot,and can meet the actual needs.