Implementation method and device of generative dialogue state tracking model

Abstract

The invention discloses a method and a device for realizing a generative dialogue state tracking model, which are used for constructing a generative model for dialogue state tracking and comprise two sub-tasks: (1) aiming at slot information multi-label classification, inputting dialogue statements into an encoder to obtain dialogue representation, and inputting the dialogue statements into a multi-label classification module to obtain logitmc corresponding to each slot; (2) generating a dialogue state task, inputting the representation of the dialogue into a decoder of the dialogue state, and generating loggen corresponding to the dialogue state by using a beam search mode; when the generative model is trained, binary cross entropy loss is calculated through logitmc and a correct slot position classification result, cross entropy loss is calculated through logitgen and a dialogue state result, and the binary cross entropy loss and the cross entropy loss are combined to train and optimize the generative model. According to the method, the single data set is used for training, the dialogue state generation accuracy is improved, the training efficiency is high, and the resource utilization rate is high.

Description

technical field [0001] The invention belongs to the technical field of computers, and relates to the technical field of natural language processing, in particular to a dialogue state tracking task, and is an implementation method and device for a generative dialogue state tracking model. Background technique [0002] Task-based dialogue enables users to easily complete tasks such as booking tickets, restaurant reservations, and customer service through natural language dialogue interactions. Behind these superficial applications, it is not difficult to find that the dialogue process is a process of continuously acquiring user needs. Only by understanding the user's needs during the dialogue process can a continuous and effective dialogue be maintained. The task-based dialogue system uses the dialogue state tracking component to track the user's needs. The dialogue state can be expressed in the form of several slot-value pairs, and each slot-value pair represents a user's req...

AI Technical Summary

Problems solved by technology

[0007] The problem to be solved by the present invention is: in the prior art, the dialog state tracking relies on predetermined slot-value pairs in the ontology, or the problem of difficulty in training multi-format data sets is inefficient, and the existing pre-trained generation model handles dialog state tracking The problem that the slot information is not fully utilized in the method

Images