CN-121980333-A - Intention recognition method, device, electronic equipment, storage medium and product

CN121980333ACN 121980333 ACN121980333 ACN 121980333ACN-121980333-A

Abstract

The disclosure provides an intention recognition method, an intention recognition device, electronic equipment, a storage medium and a product. The method comprises the steps of obtaining basic dimension characteristics in a dialogue text, wherein the basic dimension characteristics are stable characteristics of change of an involuntary graph recognition system, performing intention recognition on the basic dimension characteristics by using a recognition model to obtain a recognition result under a current intention recognition system, responding to change of the current intention recognition system, determining a classification error sample according to the recognition result, generating candidate characteristics according to error reasons of the classification error sample, determining target characteristics comprising the basic dimension characteristics and the candidate characteristics, training the recognition model based on the target characteristics, and stopping training and obtaining the intention recognition result corresponding to the target characteristics when iteration termination conditions are met.

Inventors

Che Chengfu
Jia Jingwu
XU GUANGYU
ZHENG GANG

Assignees

北京中科金得助智能科技有限公司

Dates

Publication Date: 20260505
Application Date: 20251230

Claims (10)

1. An intent recognition method, comprising: basic dimension characteristics in the dialogue text are obtained, wherein the basic dimension characteristics are stable characteristics of the change of the involuntary graph recognition system; Performing intention recognition on the basic dimension characteristics by using a recognition model to obtain a recognition result under a current intention recognition system; Responding to the change of the current intention recognition system, determining a classification error sample according to the recognition result, and generating candidate features according to the error reason of the classification error sample; Determining a target feature comprising the base dimensional feature and the candidate feature; Training the recognition model based on the target features, and stopping training when iteration termination conditions are met, so as to obtain an intention recognition result corresponding to the target features.
2. The method of claim 1, wherein the misclassification samples comprise false negative samples, false positive samples, and boundary uncertainty samples.
3. The method of claim 1, wherein generating candidate features based on the error causes of the classification error samples comprises: modifying the base dimensional feature to generate a candidate feature in response to an error cause of the classification error sample being a quality defect of the base dimensional feature; In response to the error cause of the classification error sample being insufficient coverage of the underlying dimensional feature, the target semantic feature is newly added to generate a candidate feature.
4. The method of claim 1, wherein the determining the target feature comprising the base dimensional feature and the candidate feature comprises: Verifying the validity of the candidate features; target features including the base dimensional feature and the candidate features that pass verification are determined.
5. The method of claim 4, wherein said validating the candidate feature comprises: determining the redundancy degree of the candidate feature and the basic dimension feature; and deleting the candidate feature in response to the redundancy degree exceeding a preset redundancy degree threshold.
6. The method of claim 4, wherein said validating the candidate feature comprises: determining the contribution degree of the candidate feature to intention recognition; And deleting the candidate feature in response to the contribution being below a preset contribution threshold.
7. An intent recognition device, comprising: The acquisition unit is used for acquiring basic dimension characteristics in the dialogue text, wherein the basic dimension characteristics are stable characteristics of the change of the involuntary graph recognition system; the recognition unit is used for carrying out intention recognition on the basic dimension characteristics by using a recognition model to obtain a recognition result under a current intention recognition system; The generation unit is used for responding to the change of the current intention recognition system, determining a classification error sample according to the recognition result and generating candidate features according to the error reason of the classification error sample; a determining unit configured to determine a target feature including the basic dimensional feature and the candidate feature; and the training unit is used for training the recognition model based on the target feature, and stopping training when the iteration termination condition is met, and obtaining an intention recognition result corresponding to the target feature.
8. An electronic device, comprising: at least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 6.
9. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1 to 6.
10. A computer program product comprising a computer program which, when executed by a processor, implements the method of any one of claims 1 to 6.

Description

Intention recognition method, device, electronic equipment, storage medium and product Technical Field The disclosure relates to the technical field of intention recognition, and in particular relates to an intention recognition method, an intention recognition device, electronic equipment, a storage medium and a product. Background In conversation interaction scenes such as electric marketing, customer service consultation and the like, intent recognition is a core key link supporting business decision and service optimization, and the core is mapping natural language expression of a customer into an intent recognition system. The existing intention recognition method relies on a fixed feature extraction logic and model architecture, and when the intention recognition system is adjusted due to the change of service requirements, the feature system is required to be redesigned, a large number of new samples are marked, and the model is required to be retrained in a whole amount. The method has the advantages that the problems of high feature reconstruction cost and long model adaptation period exist, the recognition precision fluctuation is easy to be caused by the logic conflict between the new features and the original features, the requirement of flexible adjustment of an intent recognition system in a business scene is difficult to be met, and the adaptation efficiency of intent recognition is limited. Disclosure of Invention The disclosure provides an intention recognition method, an intention recognition device, electronic equipment, a storage medium and a product, so as to solve the problem of low accuracy and stability of intention recognition under the condition that a category system can be frequently changed in the related art. An embodiment of a first aspect of the present disclosure proposes an intent recognition method, the method including: basic dimension characteristics in the dialogue text are obtained, wherein the basic dimension characteristics are stable characteristics of the change of the involuntary graph recognition system; Carrying out intention recognition on the basic dimension characteristics by using the recognition model to obtain a recognition result under the current intention recognition system; responding to the change of the current intention recognition system, determining a classification error sample according to a recognition result, and generating candidate features according to error reasons of the classification error sample; determining target features comprising base dimensional features and candidate features; training the recognition model based on the target features, and stopping training and obtaining the intention recognition result corresponding to the target features when the iteration termination condition is met. In one embodiment, the classification error samples include false negative samples, false positive samples, and boundary uncertainty samples. In one embodiment, generating candidate features based on the error cause of the classification error sample includes: modifying the base dimensional feature to generate a candidate feature in response to classifying an error cause of the error sample as a quality defect of the base dimensional feature; in response to the error cause of the classification error sample being insufficient coverage of the underlying dimensional feature, the target semantic feature is newly added to generate a candidate feature. In an embodiment, determining target features that include base dimensional features and candidate features includes: Verifying the validity of the candidate features; target features are determined that include the base dimensional feature and the validated candidate features. In one embodiment, validating the candidate feature includes: Determining redundancy degrees of the candidate features and the basic dimension features; and deleting the candidate features in response to the redundancy level exceeding a preset redundancy level threshold. In one embodiment, validating the candidate feature includes: Determining the contribution degree of the candidate feature to intention recognition; And deleting the candidate feature in response to the contribution being below a preset contribution threshold. An embodiment of a second aspect of the present disclosure proposes an intention recognition apparatus including: The acquisition unit is used for acquiring basic dimension characteristics in the dialogue text, wherein the basic dimension characteristics are stable characteristics of the change of the involuntary graph recognition system; the recognition unit is used for carrying out intention recognition on the basic dimension characteristics by using the recognition model to obtain a recognition result under the current intention recognition system; The generation unit is used for responding to the change of the current intention recognition system, determining a classification error sample according to