Research And Design Of Dialogue System Based On Petroleum Exploration And Production

With the increasing number of data in the field of petroleum exploration and production,the existing task-oriented dialogue systems have been difficult to accurately answer the questions asked by people,and due to the large number of professional vocabulary and unique dialogue logic in the professional field,the generality of the dialogue system applicability is not enough and intelligence is also difficult to go further.For task-oriented dialogue systems in the professional field,scholars have proposed and demonstrated a variety of methods.Although these methods have improved the performance of professional dialogue systems,they still cannot achieve the intelligence that people say.This paper studies the dialogue system in the field of petroleum exploration and production.The specific work is as follows:First,a word segmentation method in the professional field is proposed.In this paper,the key-value feature extraction method is used in the word segmentation model,and the word formation ability of each word is maximized through N-gram,which effectively improves the recognition ability of unregistered words in the professional field.Meantime,use the BERT to generate semantically rich word vectors and the Bi LSTM-CRF network to automatically assign weights,and fully consider the dependencies between labels,thereby improving the model’s professional word segmentation capabilities.It can be seen from the test that the method has a good effect in the word segmentation task in the professional field,which lays a foundation for the subsequent text processing in the field of petroleum exploration and production.Secondly,an improved intent recognition method is proposed.In order to solve the problem that text information is not fully extracted,it is proposed to take proper nouns as intent information to extract weights and embed them into the Text Rank to realize the combination of intent information and word vectors.Aiming at the problem that the input feature channels of the multi-head self-attention mechanism are not fully utilized,it is proposed to integrate the SE Net network with the multi-head self-attention mechanism,learn the importance of each input feature channel through the SE Net network,so as to improve the performance of the multi-head self-attention mechanism and improve the accuracy of intent recognition.Tests show that the proposed model performs well in the task of intent recognition.Thirdly,an improved dialogue state tracking model is proposed.The model constructs a vector space for the slot value through the Star Space,and realizes a more comprehensive feature extraction for the slot value,with this increase the recognition rate of the new slot value.The GRU is used to extract the dialogue features from the user’s current input and historical dialogue,and the Transformer is inducted to extract the features from the dialogue features and the state information of the previous round to combine,the extracted word granularity features express more comprehensive information.Combined features enable the model to be more full use of the various information that is known,so as to increase the accuracy of the dialogue state judgment.Through verification,the model has better comprehensive performance on dialogue state tracking.Finally,the dialogue system for petroleum exploration and production is implemented with NAO robot as the carrier.The text in the field of petroleum exploration and production is segmented through the improved word segmentation model,the user intent is identified by the intent recognition model,and the current dialog state is determined by the dialog state tracking model.The NAO robot is used as a terminal to communicate with the intent recognition system to form a dialogue robot.After testing,verified the accuracy of the robot in the task.