Method and apparatus for processing synthetic features, model training method, and electronic device

Patent metadata

Field	Value
Publication number	US-12430404-B2
Application number	US-202217988168-A
Country	US
Kind code	B2
Filing date	Nov 16, 2022
Priority date	Nov 18, 2021
Publication date	Sep 30, 2025
Grant date	Sep 30, 2025

Abstract

A method for processing synthetic features is provided, and includes: the synthetic features to be evaluated and original features corresponding to the synthetic features are obtained. A feature extraction is performed on the synthetic features to be evaluated based on a number S of pre-trained samples, to obtain meta features with S samples. S is a positive integer. The meta features are input into the pre-trained meta feature evaluation model for a binary classification prediction, to obtain a probability of binary classification. Quality screening is performed on the synthetic features to be evaluated according to the probability of the binary classification, to obtain second synthetic features to be evaluated. The second synthetic features are classified in a good category. The second synthetic features and original features are input into a first classifier for evaluation. classified in a poor category.

First claim

What is claimed is:

1. A method for processing synthetic features, comprising:
obtaining the synthetic features to be evaluated and original features corresponding to the synthetic features;
performing a feature extraction on the synthetic features to be evaluated based on a number S of pre-trained samples, to obtain meta features with S samples, wherein S is a positive integer;
inputting the meta features into a pre-trained meta feature evaluation model for a binary classification prediction, to obtain a probability of binary classification;
performing quality screening on the synthetic features to be evaluated according to the probability of the binary classification, to obtain a second synthetic features to be evaluated, wherein the second synthetic features are classified in a good category; and
inputting the second synthetic features and the original features into a first classifier for evaluation.

2. The method of claim 1 , wherein performing a feature extraction on the synthetic features to be evaluated based on a number S of pre-trained samples, to obtain meta features with S samples comprises:
performing the feature extraction on the synthetic features to be evaluated by adopting a minhash algorithm based on the number S of the pre-trained samples, to obtain the meta features with S samples.

3. The method of claim 1 , wherein the meta feature evaluation model and the number S of the samples are obtained by pre-training in following steps:
obtaining k synthetic features of each of then data sets, wherein n and k are both positive integers;
performing the feature extraction on the k synthetic features of each of the n data sets to obtain a meta feature set, wherein, the meta feature set comprises n*k meta features, the number of the samples in the meta feature set is s, and s is a positive integer smaller than or equal to a minimum number of the samples in the n data sets;
obtaining m original features of each of the n data sets, wherein m is a positive integer;
obtaining a classification label of each of the n*k meta features based on the k synthetic features of each of the n data sets and the m original features of each of the n data sets; and
training a second classifier based on the meta feature set and the classification label of each of the n*k meta features, obtaining model parameters and a numerical value of s, determining the numerical value of s as the number S of the pre-trained samples, and generating the meta feature evaluation model according to the model parameters.

4. The method of claim 3 , wherein obtaining a classification label of each of the n*k meta features based on the k synthetic features of each of the n data sets and the m original features of each of the n data sets comprises:
inputting m original features of each of the n data sets into the first classifier to obtain an original feature accuracy ratio of each of the n data sets;
adding the k synthetic features into the original features of each data set respectively, to obtain fusion features of the n data sets respectively;
inputting the fusion features of the n data sets into the first classifier, to obtain k synthetic feature accuracy ratios of each of the n data sets; and
obtaining the classification label of each of the n*k meta features based on the original feature accuracy ratio of each of the n data sets and the k synthetic feature accuracy ratios of each of the n data sets.

5. The method of claim 4 , wherein obtaining the classification label of each of the n*k meta features based on the original feature accuracy ratio of each of the n data sets and the k synthetic feature accuracy ratios of each of the n data sets comprises:
performing a difference calculation on the k synthetic feature accuracy ratios of each of the n data sets and the original feature accuracy ratio of each of the n data sets correspondingly, to obtain n*k difference values; and
classifying the n*k meta features based on the n*k difference values and a preset threshold, to obtain the classification label of each of the n*k meta features.

6. A method for training a meta feature evaluation model, applied to implement a synthetic feature evaluation, comprising:
obtaining k synthetic features of each of n data sets, wherein n and k are both positive integers;
performing a feature extraction on the k synthetic features of each of the n data sets, to obtain a meta feature set, wherein, the meta feature set comprises n*k meta features, a number of samples in the meta feature set is s, and s is a positive integer smaller than or equal to a minimum number of samples in the n data sets;
obtaining m original features of each of the n data sets, wherein m is a positive integer;
obtaining a classification label of each of the n*k meta features based on the k synthetic features of each of the n data sets and the m original features of each of the n data sets; and
training a second classifier based on the meta feature set and the classification label of each of the n*k meta features, obtaining model parameters and a numerical value of s, determining the numerical value of s as a number S of pre-trained samples, and generating the meta feature evaluation model according to the model parameters.

7. The method of claim 6 , wherein obtaining a classification label of each of the n*k meta features based on the k synthetic features of each of the n data sets and the m original features of each of the n data sets comprises:
inputting the m original features of each of the n data sets into a first classifier to obtain an original feature accuracy ratio of each of the n data sets;
adding the k synthetic features into the original features in each data set respectively, to obtain fusion features of the n data sets;
inputting the fusion features of the n data sets into the first classifier, to obtain k synthetic feature accuracy ratios of each of the n data sets; and
obtaining the classification label of each of the n*k meta features based on the original feature accuracy ratio of each of the n data sets and the k synthetic feature accuracy ratios of each of the n data sets.

8. The method of claim 7 , wherein obtaining the classification label of each of the n*k meta features based on the original feature accuracy ratio of each of the n data sets and the k synthetic feature accuracy ratios of each of the n data sets comprises:
performing a difference calculation on the k synthetic feature accuracy ratios of each of the n data sets and the original feature accuracy ratio of each of the n data sets correspondingly, to obtain n*k difference values; and
classifying the n*k meta features based on the n*k difference values and a preset threshold, to obtain the classification label of each of the n*k meta features.

9. A apparatus for processing synthetic features, comprising:
one or more processors;
a memory storing instructions executable by the one or more processors;
wherein the one or more processors are configured to:
obtain the synthetic features to be evaluated and original features corresponding to the synthetic features;
perform a feature extraction on the synthetic features to be evaluated based on a number S of pre-trained samples, to obtain meta features with S samples, wherein S is a positive integer;
input the meta features into a pre-trained meta feature evaluation model for a binary classification prediction, to obtain a probability of binary classification;
perform quality screening on the synthetic features to be evaluated according to the probability of the binary classification, to obtain second synthetic features to be evaluated, wherein the second synthetic features are classified in a good category; and
input the second synthetic features

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Wang Kafeng
• Xu Chengzhong
• Xiong Haoyi
• Li Xingjian
• Dou Dejing

Classifications

• G06F18/213

  Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods · CPC title

• G06F18/2415

  based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06F18/2193 Primary

  based on specific statistical tests · CPC title

• G06N20/20 Primary

  Ensemble learning · CPC title