Confidence degree assessment method and device suitable for grammar analysis

Abstract

The invention discloses a confidence degree assessment method and device suitable for grammatical analysis, wherein the method comprises the following steps of obtaining the training data, and training a translation model from a logic form to a natural language according to the training data; capturing a mapping relation between a logic form and a natural language through a translation model, obtaining natural semantics of input data and a predicted analysis result, and estimating to obtain data uncertainty; obtaining the model uncertainty of each output result through a confidence estimationalgorithm, wherein for the uncertainty of data, the posterior probability output and confidence characteristics of a basic model are connected, the abstract representation of the model uncertainty isobtained through an encoder, the conversion probability of a label is learned, and the confidence of input data is obtained. According to the method, the model prediction precision can be effectivelyimproved, and the interpretability of the prediction result can be given in an actual scene.

Description

technical field [0001] The invention relates to the technical field of grammatical analysis, in particular to a confidence evaluation method and device suitable for grammatical analysis. Background technique [0002] Today, with the rapid development of the Internet, it is the era of information and big data domination. At the same time, the demand for information visualization is increasing day by day, and data visualization provides a new method of information display. Therefore, a large number of hidden information can be discovered through data visualization technology. rules to provide a basis for decision-making. Most of the currently commonly used data visualization systems are based on a specific scene or a certain form of expression, and there is no general data visualization view generation method for information classification. [0003] Related technologies, (1) A method and system for acquiring confidence in face recognition. The method utilizes obtaining the f...

AI Technical Summary

Problems solved by technology

[0004] However, most of the existing confidence measurement methods are based on specific domains, such as images, etc.

Confidence measures established for specialized domains somewhat limit their scope of application

At the same time, most of the existing methods need another set of data storage for comparison, and finally give the uncertainty of the model results

This method does not specifically point out which part affects the effect of model confidence, and still cannot explain the reasons for the uncertainty of the model.

[0005] At the same time, most of the current confidence measurement methods are based on the posterior probability. However, due to the nonlinearity and complexity of the neural network, it is difficult for this method to accurately reflect the confidence level of the model. Therefore, a more comprehensive measurement method is needed to analyze The reasons for model uncertainty, and use model uncertainty to improve model results

[0006] In addition, most of the existing confidence estimation systems are based on statistical posterior probability methods, and do not take into account whether the input data conforms to the original data distribution. This method leads to incomplete confidence estimation, and in nonlinear environments cannot reliably give probability values ​​in

Images