Structured memory graph network model for multiple rounds of spoken language understanding

Abstract

The invention discloses a structured memory graph network model for multiple rounds of spoken language understanding, which is composed of an input coding layer, a memory coding layer, a feature aggregation layer and an output classification layer, dialogue behaviors generated by spoken language understanding tasks are used to replace texts as memory nodes for coding, and the dialogue behaviors are formatted representations containing semantic framework information. And the unstructured characters are converted into a structured triple. A graph attention network is used for replacing a recurrent neural network and an attention mechanism to achieve feature aggregation, sequence information between the attention mechanism and dialogue nodes is reserved, and model learning how to effectivelyutilize structured memory nodes is facilitated. According to the network model, the encoding dialogue behavior replaces a historical dialogue text to serve as a memory unit, original information of asemantic framework is reserved to the maximum extent, and the problems that in the prior art, noise is generated in complex occasions due to the fact that text information depends on a model, and operation efficiency is low are solved.

Description

technical field [0001] The invention belongs to the technical field of man-machine dialogue, and relates to a structured memory graph network model for multi-round spoken language comprehension. Background technique [0002] With the rapid development of various smart devices, human-computer dialogue has attracted extensive attention from academia and industry in recent years. Task-based dialogue system-related technologies have been used in many products, such as Microsoft's "Cortana" (Cortana), Apple's intelligent voice assistant Siri, etc. In a task-based dialogue system, an important module is Spoken Language Understanding (SLU), which recognizes the user's input in natural language as a semantic representation of a specific structure, including domains, intentions, slots, etc. , before being processed by other modules downstream. [0003] Most previous studies on spoken language comprehension tasks focus on single-turn dialogue scenarios. In a single-round SLU task, ...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to provide a structured memory graph network model for multi-round spoken language comprehension, which solves the problems of low operation efficiency of the above information-dependent model and noise generation in complex scenes, and reduces the time and space cost of model operation

Images