Task-based dialogue method, device and system applied to super counter and medium

Abstract

The invention provides a task-based dialogue method, device and system applied to a super counter and a medium, and the method comprises the steps: converting voice input of a user into characters, extracting private features of a business module corresponding to the characters, extracting shared features of the business module corresponding to the characters, carrying out the weighted fusion of the private features through a hybrid expert system, and carrying out the calculation of the shared features. Using an activation function to fuse the shared features and the private fusion features to obtain final fusion features, calculating global local self-attention of the final fusion features, querying a database according to the global local self-attention to obtain text response output, and converting the text response output into voice output. According to the method, the private features and the shared features of the business modules are fused, knowledge migration between the fields can be realized, and when a business scene with few marked data is encountered, related knowledge of other business modules can be migrated to the business modules, so that the business handling efficiency and the use experience are improved.

Description

technical field [0001] The present application relates to the financial field, in particular to a task-based dialogue method, device, system and medium applied to a super counter. Background technique [0002] As one of the most intelligent devices in the bank's offline outlets, the super counter, the smart teller machine, directly determines the bank's reputation and influence in the society with its business processing efficiency and customer experience. A super counter with a task-based dialogue system can The human-computer interaction mode is expanded from a single manual operation mode to a dual mode of manual operation and voice operation. [0003] Among them, the task-based dialogue system refers to the task-based human-machine dialogue system. The human-machine dialogue system interacts with humans through voice. For example, the human-machine dialogue system can be divided into a speech recognition module, a natural language processing module and a speech synthesis...

AI Technical Summary

Problems solved by technology

[0006] When applying the task-based dialogue system based on the end-to-end method, it is necessary to extract and analyze the characteristics of each field of the super counter to improve the quality of the dialogue. Specifically, each business module in the super counter can be called a field. For example, the insurance business is a field of dialogue. At present, when extracting the features of each field, the extracted data is the separate data of each field. When the data in this field is very scarce, it is difficult to effectively extract the features of this field, and Knowledge between different fields is related, so the way of extracting features alone ignores the correlation of knowledge between fields, and ignoring the correlation between fields will make it difficult for the dialogue system to transfer knowledge between fields. As a result, the dialogue system in the field of insufficient labeled data is difficult to train because it cannot learn relevant knowledge in this field from other fields, which reduces the efficiency of business processing and user experience

Images