Verification processing method and device, computer storage medium and terminal

Abstract

The invention discloses a verification processing method and device, a computer storage medium and a terminal, and the method comprises the steps: calculating the related parameter information of eachsample for a preset number of labeled samples; determining verification parameter information of each sample according to the related parameter information obtained through calculation; selecting a sample needing to be verified according to the determined verification parameter information to carry out verification processing; wherein the relevant parameter information comprises one or more of the following information: complexity, uncertainty and length. According to the embodiment of the invention, the number of samples needing to be verified is reduced, and the marking verification efficiency is improved.

Description

technical field [0001] This article involves but is not limited to information processing technology, especially a verification processing method, device, computer storage medium and terminal. Background technique [0002] With the development of information technology, the application of artificial intelligence in production and life is becoming more and more extensive. Neuro-linguistic programming (NLP) is one of the important fields in artificial intelligence, and it plays an important role in products or applications such as dialogue systems, knowledge graphs, and auxiliary judgments. NLP is mainly for text data, such as dialogue text, news, comments, referee documents, etc. Most NLP tasks, such as text classification and sequence labeling, are supervised learning, which require labeling of text categories and sequences. [0003] Annotation is mainly performed by professionals; due to the diversity of natural language itself and the possible subjective differences in t...

AI Technical Summary

Problems solved by technology

[0003] Annotation is mainly performed by professionals; due to the diversity of natural language itself and the possible subjective differences in the understanding of text and tasks by annotators, it will cause inconsistent annotation results (multiple results are acceptable) or errors

Too many inconsistencies or errors will affect the training of the model, thereby affecting the final application effect, so the labeling results need to be verified

At present, the methods of verifying the labeling results include cross-checking and random sampling inspection; among them, cross-checking is: the same data is marked by two or more people, and further review and verification of inconsistent labeling results; cross-checking The disadvantages are: 1. Two or more people labeling greatly increases the workload; 2. Consistent results do not mean that the results are correct, and it is impossible to perform special verification for data that is prone to errors or has high labeling uncertainty

The disadvantages of random sampling are: 1. The results are random, and the verification results may not reflect the overall labeling results; 2. Due to random sampling, it is not possible to perform special verification for data that is prone to errors or has high labeling uncertainty

[0004] In summary, the current method of verifying the labeling results, the verification results need to be further improved, and the verification method needs to be further improved

Images