Dialogue intention classification method and device, and nonvolatile computer storage medium

Abstract

The invention relates to a dialogue intention classification method and device, and a nonvolatile computer storage medium. The method comprises the following steps: receiving data of a conversation; extracting a first eigenvector of a sentence by utilizing a first learning network based on the data; adjusting a modulus length of the first eigenvector, so as to realize that the smaller a minimum difference is, the larger the modulus length is, and obtaining a second eigenvector. based on the second eigenvector, using a second learning network to determine probability related parameters of each known intention category, wherein the first learning network and the second learning network are jointly trained by using a first loss function serving as a measurement loss function and a second loss function representing classification loss. comparing each probability-related parameter with a threshold value to determine whether the conversation belongs to an unknown intent category and which known intent category the conversation belongs to. According to the method disclosed by the invention, when facing an open intention classification task, unknown intention categories can be effectively detected, and the classification accuracy of known intention categories can be ensured.

Description

technical field [0001] The present disclosure relates to a method and device for automatically processing and analyzing dialogues, and more particularly, to a method and device for identifying and classifying intents in dialogues. Background technique [0002] AI technologies, including conversational bots, are widely used in the automated processing and analysis of human conversations. Taking the dialogue robot as an example, it needs to effectively identify and classify the dialogue intention from the dialogue data. However, not all dialogue intents are known, and there are also unknown dialogue intentions. [0003] As shown in Table 1, dialogue robots often need to deal with the task of open intent classification, that is, among the sentences contained in the dialogue data to be analyzed, some belong to known intent categories, and some cannot identify and classify intents. The intent labels are Unknown intent. As examples of known intent categories, the intent label f...

AI Technical Summary

Problems solved by technology

[0006] Dialogue data of unknown intent categories can interfere with the recognition and classification of known intent categories, but there is currently a lack of effective methods for identifying and classifying unknown intent categories

In particular, facing the task of classifying open intentions, sentences with known and unknown intentions coexist, and it is impossible to recognize and classify both known and unknown intentions

Although the traditional supervised classification method is good for the classification of dialogue data with known intention categories, due to the lack of prior knowledge of unknown intention categories, this traditional supervised classification method will introduce false positives when classifying dialogue data with unknown intention categories. Positive errors, and there is currently a lack of technical means to effectively use the dialogue data of known intent categories to generalize to the recognition and classification of unknown intent categories

Images