Model training method, text classification method and related devices

Abstract

The invention provides a model training method, a text classification method and related devices, and relates to the technical field of machine learning. According to the invention, when the inaccurate supervised learning is carried out on the neural network model by utilizing the training sample with accurate label labeling and the training sample with inaccurate label labeling, a respective corresponding proportion parameter for each training sample is determined; a weight parameter vector representing the accurate labeling probability of all training sample labels is obtained; the feature vectors corresponding to all the training samples are learned by using the weight parameter vectors; proportional output vectors are obtained, a vector based on the proportion is outputted. Compared with the prior art, the model parameters of the neural network model are updated, so that the neural network model can learn more accurate sample information marked by the label when learning is not accurately supervised. The inaccurate sample information marked by the learning label is reduced. The training effect of the neural network model when learning is not accurately supervised is improved.

Description

Technical field [0001] This application relates to the field of machine learning technology, and specifically to a model training method, a text classification method and related devices. Background technique [0002] Supervised learning refers to training a neural network model using labeled training samples, where the labels labeled by the training samples represent the true output of the training samples. [0003] However, in the actual training process, because labeling training samples requires a lot of manpower, material resources and other costs, there are always a small number of labeled training samples. Generally, more training samples are not labeled with labels, or labeled with thick labels. The granular labels or the labeled labels are not necessarily all true values; for this reason, weakly supervised learning is generally used to overcome the situation that there are fewer training samples labeled with true value labels. [0004] However, due to weakly supervised lear...

AI Technical Summary

Problems solved by technology

[0003] However, in the actual training process, since it takes a lot of manpower and material resources to label the training samples, there are always a small number of labeled training samples. Generally, more training samples are not labeled with labels, or are labeled with rough labels. Granular labels, or labeled labels are not necessarily all true values; for this reason, weakly supervised learning is generally used to overcome the fact that there are few training samples labeled with true value labels

[0004] However, due to weakly supervised learning, such as inaccurate supervision (inaccurate supervision) learning, since there are a large number of training samples during training, the labels marked may not be true, and the model cannot learn the correct features of the training samples well, thus The training effect of the model is often poor when it leads to inaccurate supervised learning

Images