Transfer Learning from LDA to BiLSTM-CNN for Offensive Language Detection in Twitter

TL;DR

This paper explores transfer learning strategies to enhance offensive language detection on German Twitter data using a BiLSTM-CNN model, comparing supervised, weakly-supervised, and unsupervised approaches, and addressing catastrophic forgetting.

Contribution

It introduces a novel transfer learning framework combining unsupervised topic modeling with thematic user data for improved offensive language classification.

Findings

Transfer learning generally improves detection performance.

Unsupervised topic clustering combined with user data yields best results.

Mitigation strategies for catastrophic forgetting are effective.

Abstract

We investigate different strategies for automatic offensive language classification on German Twitter data. For this, we employ a sequentially combined BiLSTM-CNN neural network. Based on this model, three transfer learning tasks to improve the classification performance with background knowledge are tested. We compare 1. Supervised category transfer: social media data annotated with near-offensive language categories, 2. Weakly-supervised category transfer: tweets annotated with emojis they contain, 3. Unsupervised category transfer: tweets annotated with topic clusters obtained by Latent Dirichlet Allocation (LDA). Further, we investigate the effect of three different strategies to mitigate negative effects of 'catastrophic forgetting' during transfer learning. Our results indicate that transfer learning in general improves offensive language detection. Best results are achieved from pre-training our model on the unsupervised topic clustering of tweets in combination with thematic user cluster information.