LIIR at SemEval-2021 task 6: Detection of Persuasion Techniques In Texts and Images using CLIP features

TL;DR

This paper presents a multimodal approach using CLIP features and data augmentation for detecting persuasion techniques in memes, achieving competitive rankings in SemEval-2021 task 6.

Contribution

It introduces a novel combination of pretrained CLIP models with chained classifiers and data augmentation for improved persuasion detection.

Findings

Achieved 8th place in F1-micro ranking

Achieved 9th place in F1-macro ranking

Demonstrated effectiveness of multimodal features and data augmentation

Abstract

We describe our approach for SemEval-2021 task 6 on detection of persuasion techniques in multimodal content (memes). Our system combines pretrained multimodal models (CLIP) and chained classifiers. Also, we propose to enrich the data by a data augmentation technique. Our submission achieves a rank of 8/16 in terms of F1-micro and 9/16 with F1-macro on the test set.