Training data classification method and device, equipment and storage medium

In the process of collecting incremental pre-training corpus, multiple pre-trained models are used to perform classification prediction and annotation strategy calculation, and the target classification model is trained, which solves the problems of low manual screening efficiency and high cost of automation tools in the existing technology, and realizes efficient and low-cost incremental pre-training corpus screening.

CN119719985APending Publication Date: 2025-03-28BEIJING KINGSOFT CLOUD NETWORK TECH CO LTD +1
0 Cites 0 Cited by

Patent Information

Application Number
CN202411845533.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-13
Publication Date
2025-03-28

AI Technical Summary

Technical Problem

In the process of collecting incremental pre-training corpus, manual screening is inefficient and costly, automation tools rely on large models for classification, and computing resources and cost are high.

Method used

By obtaining multiple pretraining data related to the application scenario, dividing them into evaluation sets and training sets, using multiple pretrained models to classify and predict under the setting annotation standards, calculating the annotation strategy value, determining a high-quality pretrained model, classifying and predicting the training set, and training the target classification model based on this.

Benefits of technology

It reduces the cost of screening training data, improves screening efficiency, ensures classification accuracy, saves labor costs, and reduces the overhead of using large models.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to the technical field of classification, in particular to a training data classification method and device, equipment and a storage medium, and the method comprises the steps: obtaining a plurality of pieces of pre-training data related to an application scene, and dividing the plurality of pieces of pre-training data into an evaluation set and a training set; performing classification prediction on the evaluation set under a set marking standard through a plurality of pre-training models, and outputting a first prediction category; calculating an annotation strategy value based on the first prediction category and the annotation category; determining at least one first pre-training model in a plurality of pre-training models according to the labeling strategy value; performing classification prediction on the training set through at least one first pre-training model, and determining a second prediction category; and training the initial classification model according to the training set and the second prediction category to obtain a trained target classification model. According to the method provided by the invention, classification labeling of the training data is completed, the classification cost is reduced and the screening efficiency is improved under the condition that the classification precision is ensured.
Need to check novelty before this filing date? Find Prior Art