FPAI at SemEval-2021 Task 6: BERT-MRC for Propaganda Techniques Detection

Abstract

The objective of subtask 2 of SemEval-2021 Task 6 is to identify techniques used together with the span(s) of text covered by each technique. This paper describes the system and model we developed for the task. We first propose a pipeline system to identify spans, then to classify the technique in the input sequence. But it severely suffers from handling the overlapping in nested span. Then we propose to formulize the task as a question answering task by machine reading comprehension (MRC) framework which achieves a better result compared to the pipeline method. Moreover, data augmentation and loss design techniques are also explored to alleviate the problem of data sparsity and imbalance. Finally, we attain the 3 rd place in the final evaluation phase.

PeerDA: Data Augmentation via Modeling Peer Relation for Span Identification Tasks

“…(2) Self-training is to first train a model with manually labeled data, then use the model to automatically label unlabeled data, and finally leverage the manually and automatically labeled data to enhance itself (Xie et al, 2019(Xie et al, , 2020. It shows promising results in many SpanID tasks, including NER (Wang et al, 2020), propaganda detection (Hou et al, 2021) (Seo et al, 2016;Chen et al, 2017), while recent trends have shown great advantages of formulating NLP tasks as MRC problems. In the con- (Li et al, 2019a), event detection ), and summarization (McCann et al, 2018 are also reported to benefit from the MRC paradigm.…”

“…Therefore, we only compare with MRC. For So-cial21, we compare with top three approaches in its leaderboard, namely, Volta (Gupta et al, 2021), HOMADOS (Kaczyński and Przybyła, 2021), and TeamFPAI (Hou et al, 2021).…”