DMCA
Hierarchical Multi-Label Classification of Social Text Streams
| Citations: | 3 - 0 self |
BibTeX
@MISC{Ren_hierarchicalmulti-label,
author = {Zhaochun Ren and Maria-hendrike Peetz and Shangsong Liang and Willemijn Van Dolen and Maarten De Rijke},
title = {Hierarchical Multi-Label Classification of Social Text Streams},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Hierarchical multi-label classification assigns a document to mul-tiple hierarchical classes. In this paper we focus on hierarchical multi-label classification of social text streams. Concept drift, com-plicated relations among classes, and the limited length of docu-ments in social text streams make this a challenging problem. Our approach includes three core ingredients: short document expan-sion, time-aware topic tracking, and chunk-based structural learn-ing. We extend each short document in social text streams to a more comprehensive representation via state-of-the-art entity link-ing and sentence ranking strategies. From documents extended in this manner, we infer dynamic probabilistic distributions over top-ics by dividing topics into dynamic “global ” topics and “local ” top-ics. For the third and final phase we propose a chunk-based struc-tural optimization strategy to classify each document into multi-ple classes. Extensive experiments conducted on a large real-world dataset show the effectiveness of our proposed method for hierar-chical multi-label classification of social text streams.
Keyphrases
social text stream hierarchical multi-label classification challenging problem local top-ics sentence ranking strategy dynamic probabilistic distribution final phase short document expan-sion hierar-chical multi-label classification dynamic global topic mul-tiple hierarchical class extensive experiment state-of-the-art entity time-aware topic tracking large real-world dataset concept drift limited length com-plicated relation chunk-based structural learn-ing multi-ple class short document core ingredient chunk-based struc-tural optimization strategy comprehensive representation
