A Knowledge Graph Based Dialogue Command Understanding Method

Abstract

The present invention provides a method for understanding dialogue commands based on knowledge graphs. The method for understanding dialogue commands based on knowledge graphs is based on a preset domain knowledge set, correspondingly constructs a domain knowledge spectrum library, and converts the input dialogue commands into several A candidate list of candidate matching sequences, and then convert each candidate matching sequence in the candidate list into a knowledge map vector through a preset encoding method, and thus form a dialog command vector about the input dialog command, and finally based on the The dialogue command vector is used to train the deep learning neural network model, and the semantic understanding operation of the dialogue command is performed through the trained deep learning neural network model.

Description

technical field [0001] The invention relates to the technical field of speech recognition, in particular to a method for understanding dialog commands based on knowledge graphs. Background technique [0002] In the dialogue system, the understanding of dialogue commands refers to receiving and parsing the utterance from the user, and identifying the intention of the utterance, so as to extract the relevant parameters (also called slots) in the utterance, so that the dialogue system can understand the user utterance semantic understanding. There are mainly two ways to understand user utterance commands in existing dialogue systems. One is to use grammatical and semantic rules to extract the semantic information in the utterance command text, and the other is to convert utterance intention recognition and slot extraction into Text classification and sequence labeling tasks for machine learning, and use corpus training models to predict semantic information in utterance comman...

AI Technical Summary

Problems solved by technology

[0003] Although the existing understanding of user utterance commands has a good semantic analysis effect for utterance commands with regular distribution, its own text fuzzy matching ability is poor, and it is difficult to correctly understand the widely existing users' free and arbitrary expressions. Accurate Semantic Parsing of Discourse Text

In actual application scenarios, basically all users issue speech commands according to their daily speaking habits, and these speech commands formed based on different users’ speaking habits do not strictly conform to the pre-set speech rules of the machine, which is serious. This greatly limits the applicability of the existing understanding of user utterance commands to different application scenarios

In addition, the existing machine learning model for understanding user utterance commands is to automatically learn the intrinsic relationship between the prediction target and the natural language text from a large number of training corpora, or to improve its semantics by simply adding dictionary information in different fields. Analytical ability, but this makes it difficult to embed complex domain map knowledge in the existing understanding of user utterance commands. If the understanding of user utterance commands is separated from the support of specific domain knowledge, it is difficult to learn from vague or irregular Extracting the correct semantics from utterance commands severely limits the improvement of the level of semantic understanding of user utterance commands

Images