Transformer-based Context-aware Sarcasm Detection in Conversation Threads from Social Media
This work addresses sarcasm detection for social media analysis, showing incremental improvements over existing methods.
The paper tackles sarcasm detection in social media conversations by using a transformer-based model that incorporates context from entire threads, achieving improvements of 3.1% and 7.0% over baselines on Twitter and Reddit datasets with F1-scores of 79.0% and 75.0%, respectively.
We present a transformer-based sarcasm detection model that accounts for the context from the entire conversation thread for more robust predictions. Our model uses deep transformer layers to perform multi-head attentions among the target utterance and the relevant context in the thread. The context-aware models are evaluated on two datasets from social media, Twitter and Reddit, and show 3.1% and 7.0% improvements over their baselines. Our best models give the F1-scores of 79.0% and 75.0% for the Twitter and Reddit datasets respectively, becoming one of the highest performing systems among 36 participants in this shared task.