Multi-task Learning of Pairwise Sequence Classification Tasks Over Disparate Label Spaces

TL;DR

This paper introduces a novel multi-task learning approach that combines semi-supervised learning by creating a joint embedding space for different label sets, improving performance on sequence classification tasks.

Contribution

It proposes a new method for joint embedding of disparate label spaces and transfer functions, enabling effective multi-task learning across varied datasets.

Findings

Outperforms strong single and multi-task baselines

Achieves state-of-the-art in topic-based sentiment analysis

Effectively leverages unlabeled data and auxiliary datasets

Abstract

We combine multi-task learning and semi-supervised learning by inducing a joint embedding space between disparate label spaces and learning transfer functions between label embeddings, enabling us to jointly leverage unlabelled data and auxiliary, annotated datasets. We evaluate our approach on a variety of sequence classification tasks with disparate label spaces. We outperform strong single and multi-task baselines and achieve a new state-of-the-art for topic-based sentiment analysis.