Universally configurable, multi-model compatible, and domain-transferable multi-round dialogue system

Abstract

The invention discloses a multi-round dialog system that is universally configurable, compatible with multiple models, and domain transferable. Understand the semantic encoding of user sentences, and perform intent recognition and joint extraction of slot-value pairs by replacing a compatible dialogue understanding model; the dialogue state tracking module is used to obtain the dialogue state information of the current multiple rounds of history; the policy learning module is used to pass The dialogue strategy model is used for action decision-making; the dialogue generation module is used for dialogue generation according to the corresponding actions; the online training module is used as a user simulator for developers, and iteratively updates the model online and generates available data. The system's overall architecture modules are common and adapt to different application scenarios. The parameters of each field are common and configurable, and corpus can be accepted at will. The dialogue understanding and dialogue strategy models can be flexibly expanded and replaced.

Description

technical field [0001] The present invention relates to the technical field of information technology and data business, in particular to a multi-round dialogue system that is universal and configurable, compatible with multiple models and can be transferred in fields. Background technique [0002] At present, the task-oriented multi-round dialogue system is mainly based on the pipeline structure, and its purpose is generally to complete the design of a series of internal modules for tasks in a specific field. , cannot be applied in multiple fields, the fixed business rules are mixed with the system, the specified fields cannot be changed, and the internal model is unique and cannot be updated, the coordination between modules is fragile, changing the internal model structure requires a lot of labor costs, and its system fields Immobilization, application singleness, and model solidification restrict its application and development in different fields. Versatility, multi-co...

AI Technical Summary

Problems solved by technology

Moreover, its model is fixed, and it is impossible to continuously iteratively update the existing model, resulting in the inability to improve the model effect. The change requires a lot of logic transformation, and the fields need to be redefined, which consumes a lot of labor costs.

Images