Method and system for constructing multi-level classification model

Abstract

The invention provides a method and a system for constructing a multi-level classification model. The method comprises the steps of inputting marked data samples, generating an initial multi-level classification model by the marked data samples and optimizing the initial multi-level classification model, wherein the optimizing step can comprise the process of adjusting the initial multi-level classification model based on the global level space among all levels. In another embodiment, the optimizing step can comprise the step of adjusting the initial multi-level classification model based on the order relation among the data samples; and in another embodiment, the method is expanded and applied to the half-supervised learning situation, and can enable the initial multi-level classification model to be more smooth based on the similarity among the data samples and the similarity among the levels through referring to marked data and not-marked data.

Description

technical field [0001] The present invention relates generally to information classification, and more particularly to multi-class classification and multi-level classification for classifying information samples into multiple classes. More specifically, the present invention relates to methods and systems for building multi-class classification models. Background technique [0002] In traditional multi-category information classification methods, categories are usually independent and disordered. For example, in the classification of news, the categories of news may include politics, economy, military affairs, science and so on. [0003] In real life, however, there is another special kind of multi-class problem. That is, there is an orderly relationship between the categories and a smooth distribution. Such classification problems are known as multi-class classification problems. In multiclass classification problems, information samples are classified into different c...

AI Technical Summary

Problems solved by technology

In the above-mentioned ordinal regression model optimization method of prior art 1, only the order relationship between limited adjacent levels is considered

Also, the output level labels in an ordinal regression model are discrete and cannot be measured continuously

Furthermore, no existing multilevel or ordinal regression methods can be applied in semi-supervised learning situations such as prior art 2

Images