Dialogue state generation method based on hierarchical multi-head interaction attention

Abstract

The invention belongs to the technical field of task type dialogues, particularly relates to a dialogue state generation method based on hierarchical multi-head interaction attention, and aims to solve the problems of low precision and accuracy, high cost and low efficiency in the prior art. The method comprises the following steps: performing dialogue text preprocessing based on a text dictionary; independently encoding each sentence through an encoder to obtain context representation of the dialogue text; applying a self-attention mechanism to decoder input to obtain a decoder input vector at the current moment; applying a multi-head interactive attention mechanism to fuse the context representations of the word level and the sentence level, and obtaining a context vector representationof the dialogue text at the current moment; and in combination with the decoder input vector at the current moment, obtaining an entity and a state as a dialogue state of the dialogue text through nonlinear mapping. According to the method, a very good effect can be achieved under the condition of no word-level annotation information, so that the data annotation cost is saved, and the accuracy andprecision of the model are also improved.

Description

technical field [0001] The invention belongs to the technical field of task-based dialogue, and in particular relates to a dialogue state generation method based on hierarchical multi-head interactive attention. Background technique [0002] A task-based dialogue system is a human-computer interaction system that assists users to complete a specific task in a specific field through natural language interaction. At present, task-based dialogue systems are in great demand in various vertical fields, especially in the medical field. In the medical dialogue system, the analysis and understanding of the user dialogue text is the first step in constructing the medical dialogue system. First, it is necessary to identify disease-related entities such as symptoms, examinations, and drugs that appear in user texts and dialogue history, and then infer the state information of these entities. [0003] In medical dialogue texts, the grammatical structure is not standardized and the phe...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the above-mentioned problems in the prior art, that is, the existing end-to-end method ignores the characteristics of multi-round interaction of dialogue text information, the precision and accuracy of dialogue state generation is low, and it relies heavily on text-level annotation information, thus The problem of high cost and low efficiency, the present invention provides a dialogue state generation method based on hierarchical multi-head interactive attention, the method includes:

Images