1213Li at SemEval-2021 Task 6: Detection of Propaganda with Multi-modal Attention and Pre-trained Models

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 1032–1036 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 1032 1213Li at SemEval-2021 Task 6: Detection of Propaganda with Multi-modal Attention and Pre-trained Models Peiguang Li 1,2,3,4, Xuan Li 1,2,3,4,∗, Xian Sun 1,2,3,4,† 1 Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100190, China 2 Key Laboratory of Network Information System Technology (NIST), Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100190, China 3 University of Chinese Academy of Sciences, Beijing 100190, China 4 School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100190, China {lipeiguang17, lixuan173}@mails.ucas.ac.cn, sunxian@aircas.ac.cn Abstract This paper presents the solution proposed by the 1213Li team for subtask 3 in SemEval- 2021 Task 6: identifying the multiple persua- sion techniques used in the multi-modal con- tent of the meme. We explored various ap- proaches in feature extraction and the detec- tion of persuasion labels. Our final model em- ploys pre-trained models including RoBERTa and ResNet-50 as a feature extractor for texts and images, respectively, and adopts a label embedding layer with multi-modal attention mechanism to measure the similarity of la- bels with the multi-modal information and fuse features for label prediction. Our pro- posed method outperforms the provided base- line method and achieves 3rd out of 16 partic- ipants with 0.54860/0.22830 for Micro/Macro F1 scores. 1 Introduction The development of the Internet and Information Technology promotes the generation and dissemi- nation of information, but also fuels the prolifera- tion of disinformation. As one of the most popular types of content in disinformation, memes attract readers easily and brought further challenges to the detection of disinformation (Martino et al., 2020; Dimitrov et al., 2021). Specifically, memes employ a number of tech- niques to influence users, which can be divided into the use of logical fallacies and appealing to the emotions of the audience (Dimitrov et al., 2021). In practice, the former misuses logical rules to dis- guise wrong conclusions as correct and objective, ∗Co-author. †Corresponding author. ‡https://www.163.com/dy/article/ F0HKK63D0511EPAO.html §http://www.zhujia120.com/fenxi/ 202103/318716.html ¶https://www.163.com/dy/article/ G56M8U7R05521HYB.html Figure 1: Examples of multi-modal samples, we re- write the sentences on our own and collect the images from‡, §, ¶, respectively. The first two rows illustrate the visual and the textual content, and in the last row, each line reveals the label (techniques) of the sample. while the latter utilizes emotional language to in- duce the audience to agree with the speaker emo- tionally and prevent their rational analysis of the argumentation. Identifying the techniques used in memes con- tributes to the understanding of user-generated con- tent and further helps to the detection of disinforma- tion. The subtask 3 of SemEval-2021 Task 6 (Dim- itrov et al., 2021) is organized to stimulate the study of computational methods to detect persuasion tech- niques in memes that inhere in texts and images. As shown in Figure 1, each sample consists of a set of textual sentences and an attached image. According to the task description (Dimitrov …

Content truncated to 500 words