Three-way decision active learning method taking neighborhood entropy as query strategy

Abstract

The invention discloses a three-way decision active learning method taking neighborhood entropy as a query strategy. The method comprises the following steps: training a classifier by using a marked data set; classifying the test set by using the trained classifier and recording a classification result of the test set; calculating neighborhood entropies of all the unmarked data, and dividing the unmarked data into a positive domain, a boundary domain and a negative domain according to the size of the neighborhood entropies; respectively processing the data of different areas; selecting a part of most valuable unmarked data, and marking the unmarked data by a human expert or an annotation device; after marking, adding a marked data set and using the marked data set for the next training of the classifier; and executing the above processes in a loop iteration manner until a preset condition or an expected evaluation standard is reached, and stopping learning. According to the method, a small amount of most valuable data can be selected and marked, selection of redundant data and data which contribute little to classification performance is avoided, and meanwhile, the cost for marking a large amount of unmarked data can be reduced.

Description

technical field [0001] The invention belongs to the technical field of data mining and information processing, and in particular relates to a three-way decision-making active learning method using neighborhood entropy as a query strategy. Background technique [0002] In real life, unlabeled data is abundant and easy to obtain. These unlabeled data contain a lot of effective information. However, it takes a lot of manpower and material resources to manually extract useful information from a large amount of unlabeled data. Therefore, how never Mining the most effective information from labeled data has become a major research hotspot. Active learning (Active Learning) is one of the widely used machine learning methods, which aims to reduce the amount of labeled data required, that is, it only needs to use less training data to train a classifier with better performance. Active learning uses an appropriate query strategy to iteratively select the most valuable unlabeled data ...

AI Technical Summary

Problems solved by technology

At the same time, the neighborhood entropy is used as the decision function to map the unlabeled data set to three regions, and then the data in different regions are processed separately to solve the problem that the data in different regions have different values.

Images