Hierarchical Feature Selection Algorithm Combined with Category Information Constraints

Feature selection faces great challenges due to the expansion of the labelling space and inevitably noisy data. Due to the large number of classes, planar feature selection methods are unable to access compact and excellent feature subsets, while in real ownership scenarios, many practical tasks usually organize the classes in a coarse-to-fine manner, which can be used as a partitioning strategy to solve the related problems. Hierarchical classification is a typical classification method of partitioning strategy, It is a special type of multi-token classification, in which an instance can be associated with multiple tokens, each of which has its own rich semantic descriptions, and thus problems such as semantic ambiguity among category tokens are unavoidable. Aiming at the mentioned problems, in this paper, we propose a hierarchical feature selection method combined with category information constraints.