BOOMER — An algorithm for learning gradient boosted multi-label classification rules

Multi-label classification is concerned with the assignment of sets of labels to individual data points. Due to its diverse real-world applications, e.g., the annotation of text documents with topics, it has become a well-established field of machine learning research. Compared to traditional classi...

Full description

Saved in:
Bibliographic Details
Published in:Software impacts Vol. 10; p. 100137
Main Author: Rapp, Michael
Format: Journal Article
Language:English
Published: Elsevier B.V 01.11.2021
Subjects:
ISSN:2665-9638, 2665-9638
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Multi-label classification is concerned with the assignment of sets of labels to individual data points. Due to its diverse real-world applications, e.g., the annotation of text documents with topics, it has become a well-established field of machine learning research. Compared to traditional classification, where classes are mutually exclusive, multi-label classification comes with interesting challenges, most prominently the requirement to take dependencies between labels into account. In this work, we present a modular and customizable implementation of BOOMER – an algorithm for learning gradient boosted multi-label classification rules – that can flexibly be adjusted to different use cases and requirements. •BOOMER is an algorithm for learning gradient boosted multi-label classification rules.•The goal of multi-label classification is the automatic assignment of sets of labels to individual data points.•BOOMER enables to optimize decomposable and non-decomposable loss functions.•The implementation incorporates several optimizations and approximation techniques to be able to deal with large datasets.•Gradient-based Label Binning can be used to form groups of similar labels.
ISSN:2665-9638
2665-9638
DOI:10.1016/j.simpa.2021.100137