Research On Task-based Dialogue Oriented To Customer Service Assistance Domain

In recent years,with the rapid development of artificial intelligence technology,task-based dialogue research based on natural language has attracted much attention from academia and industry.Current customer service dialogue is mainly based on manual participation,so how to improve the service experience and implementation of intelligent customer service is the focus of this paper.Adding natural language dialogues to the field of customer service dialogues can effectively reduce manpower consumption and improve the efficiency of customer service dialogues.The main research directions of Task-based dialogue can be divided into two types:retrieval method and generative method.The retrieval method performs smoothly in the limited field,and the content of the generative method is more innovative.Because in-depth learning can effectively extract the mutual information between semantic features and context,it provides a new research idea for this paper.This paper mainly focuses on the task-based dialogue research in the field of customer service assistance,focusing on the following three directions:Research on Retrieval Dialogue Oriented to Customer Service Assistance Domain.Traditional dialog research is mainly based on knowledge base retrieval and response matching.This paper implements three retrieval dialog models: WMD,BM25 and SMN.The experimental comparison shows that the accuracy of SMN model with deep semantic understanding is close to 40%,which is obviously superior to the other two.This shows that deep learning method is conducive to extracting semantic feature information.Research on Generative Dialogue Oriented to Customer Service Assistance Domain.Cyclic neural network has the characteristics of memory,parameter sharing,and is good at processing serialized data information.This paper implements the baseline model of seq2 seq,and uses LSTM and Bi-LSTM to improve the model coding and enrich the expression of semantic information.On this basis,attention mechanism is introduced to enhance the decoding ability of the model.The improved experimental model achieves68.3% accuracy,which is better than the baseline model.This shows that the deep learning method has a significant effect on the understanding of semantic information,but the experimental accuracy still needs to be improved.To sum up,the combination of retrieval and generation is used to conduct experimental dialogue research.In this paper,the keywords of retrieval information and retrieval information are added to the coding and decoding parts of the generated model to enrich the coding of model information and improve the decoding results.The optimalexperimental results are 77.41%.In order to increase the landing feasibility of the dialogue model,task-based dialogue was added on this basis,which was combined with search-based dialogue and generative dialogue based on certain strategies.Finally,the test results reached 88.71%.Finally,the integrated model combines the advantages of retrieval and generation,and greatly improves the efficiency of dialogue and service experience.Experiments show that the scheme is more practicable.