Research On Spoken Language Understanding Approach Based On Deep Learning

The task-oriented human-machine dialogue system is the current research hotspot in the field of natural language processing,and the development of related technologies has been receiving extensive attention from academia and industry in recent years.Spoken language understanding is an important part of the task-oriented dialogue system,which aims to transform the utterance input by the user in the form of natural language into a structured semantic representation,including domains,intents,and slots.Although the performance of the technical models proposed by the spoken language understanding research has been gradually improving in recent years,due to the diverse forms of dialogue systems,the existing spoken language understanding models still have difficulties such as large task differences and weak generalization capabilities.In response to the above problems,this article starts from three different forms of dialogue systems,and studies the application scenarios and improvements of different spoken language understanding technologies.The main contributions of this article are as follows:Firstly,Aiming at the problem that the subtasks of single-turn spoken language understanding cannot perform the interaction of explicit features,we propose a typed iteration approach.This approach uses a typed abstraction mechanism to abstract the encoded information of the slot filling task,which is used to enhance the performance of the intent detection task.In addition,the method realizes two-way interaction of encoded information through a iteration mechanism,and mitigating the negative impact of error propagation.Experiments on two public English data sets(ATIS and SNIPS)show that the performance of the method proposed in this paper is significantly improved on multiple indicators.Secondly,Aiming at the task of multi-intent spoken language understanding,this paper proposes an independent dynamic interactive network.In the past research on spoken language understanding,an important assumption was that the user only expressed one intent in each round of dialogue.However,in the real worlds,users may express multiple intents in one sentence.For this special scenario,we propose an adaptive interactive network.By using the independent dynamic interaction layer based on the attention mechanism,the network captures semantic information with different meanings for each token.Experimental results on multiple multi-intent spoken language understanding datasets such as DSTC4 show that the network proposed in this paper has a significant improvement in the performance of multi-intent detection scenarios.Thirdly,For multi-turn dialogue spoken language understanding tasks,Previous studies exploited the historical dialogue information by attention-based graph structure simulation,but these methods cannot explicitly take advantage of the structure of the dialogue state.In addition,how to generate complex format dialogue states also brings challenges to research.In this paper,we propose a State Memory Graph Network(SMGN).The network saves historical information through the state memory graph,and uses the graph to interact with the current dialogue.We also implement a complex dialogue state generation method based on state memory graph.Experimental results on multiple open multi-turn dialogue datasets(Chinese dataset Cross WOZ,English datasets Multi WOZ 2.0 and Multi WOZ 2.1)show that the method proposed in this paper can effectively improve the joint accuracy.