Research On Key Technologies Of Intention Detection And Slot Filling

Intent detection tasks and slot filling tasks are basic tasks in the field of natural language understanding.Intent detection requires identifying the intent category of the text sequence,while slot filling requires extracting the slot value from the text sequence according to the intent category to further clarify the specific content of the text sequence.On the one hand,the research on intention recognition tasks and slot filling tasks can effectively promote the development of natural language understanding;on the other hand,the research on intention detection tasks and slot filling tasks is of great significance to the research of specific downstream tasks such as dialogue systems.The current deep learning models have achieved good results in both intent detection and slot filling,but the recognition effect of colloquial sentences that omits many key semantic information still needs to be improved.In this paper,external information is introduced into the intention recognition model to further enhance the model's semantic modeling effect on short sentences.At the same time,a semantic enhancement module is added to the intention recognition and slot filling combined model to further improve the performance of the combined model.This article mainly carried out the following three research tasks:(1)The Chinese intention recognition and slot filling data set are labeled,and the knowledge graph highly related to the content of the data set is constructed.Aiming at the problem of the lack of Chinese intention recognition and slot filling data sets currently used in academia,this paper has completed the work of intention recognition and slot information labeling on the short question corpus collected in major forum communities;at the same time,Based on the knowledge base related to the question corpus,a knowledge graph that is highly related to the content of the data set is constructed.(2)Introducing external information into intention recognition model.Due to the key information omission and colloquialization of the short question data set,the performance effect of the deep learning model needs to be improved.In this paper,knowledge graph is introduced into the pre training language model as external information to improve the performance of the pre training language model on the short question data set.(3)Introducing instruction information into the join model of intent detection and slot filling.In this paper,firstly,the entity semantic relationship in the knowledge graph is integrated into the join model of intention recognition and slot filling,and a good performance effect is achieved on the experimental data set;Finally,through the analysis of error examples,the joint model integrated with entity semantic relationship is further modified,and the semantic enhancement module is added.The comparative experiments show that the introduction of instruction information can further improve the performance of the joint model.