NLyticsFKIE at SemEval-2021 Task 6: Detection of Persuasion Techniques In Texts And Images

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 1037–1044 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 1037 NLyticsFKIE at SemEval-2021 Task 6: Detection of Persuasion Techniques In Texts And Images Albert Pritzkau Fraunhofer FKIE / Wachtberg, Germany albert.pritzkau@fkie.fraunhofer.de Abstract The following system description presents our approach to the detection of persuasion tech- niques in texts and images. The given task has been framed as a multi-label classification problem with the different techniques serving as class labels. The multi-label classification problem is one in which a list of target vari- ables such as our class labels is associated with every input chunk and assumes that a docu- ment can simultaneously and independently be assigned to multiple labels or classes. In order to assign class labels to the given memes, we opted for RoBERTa (A Robustly Optimized BERT Pretraining Approach) as a neural network architecture for token and se- quence classification. Starting off with a pre- trained model for language representation we fine-tuned this model on the given classifica- tion task with the provided annotated data in supervised training steps. To incorporate im- age features in the multi-modal setting, we rely on the pre-trained VGG-16 model archi- tecture. 1 Introduction Social networks provide opportunities to conduct disinformation campaigns for organizations as well as individual actors. The proliferation of disinfor- mation online, has given rise to a lot of research on automatic fake news detection. SemEval-2021 Task 6 considers disinformation as a communica- tion phenomenon. By detecting the use of various persuasion techniques in communication, it takes into account not only the content but also how a subject matter is communicated. The goal of the shared task is to build models for identifying such techniques in the textual content of a meme only (two subtasks) and in a multimodal setting in which both the textual and the visual content are to be analysed together (one subtask). The shared task defines the following subtasks: Subtask 1 Given only the “textual content” of a meme, identify which of the 20 tech- niques are used in it. This is a multil- abel classification problem. Subtask 2 Given only the “textual content” of a meme, identify which of the 20 tech- niques are used in it together with the span(s) of text covered by each tech- nique. This is a multilabel sequence tagging task. The task is the combina- tion of the two subtasks of the SemEval 2020 task 11 on ”detecting propaganda techniques in news articles”. Note that subtask 1 is a simplified version of sub- task 2 in which the spans covered by each technique is not supposed to be provided. Subtask 3 Given a meme, identify which of the 22 techniques are used both in the textual and visual content of the meme (multi- modal task). This is a multilabel classi- fication problem. In this work, we covered our approach on both technique classification (TC) tasks (Subtask 1 and Subtask 3) detecting the type of communication technique …

Content truncated to 500 words