NLFIIT at SemEval-2020 Task 11: Neural Network Architectures for Detection of Propaganda Techniques in News Articles

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 1771–1778 Barcelona, Spain (Online), December 12, 2020. 1771 NLFIIT at SemEval-2020 Task 11: Neural Network Architectures for Detection of Propaganda Techniques in News Articles Matej Martinkovic, Samuel Pecar, Marian Simko Slovak University of Technology in Bratislava Faculty of Informatics and Information Technologies Ilkovicova 2, 842 16 Bratislava, Slovakia {xmartinkovicm1, samuel.pecar, marian.simko}@stuba.sk Abstract Since propaganda became more common technique in news, it is very important to look for possibilities of its automatic detection. In this paper, we present neural model architecture submitted to the SemEval-2020 Task 11 competition: ”Detection of Propaganda Techniques in News Articles”. We participated in both subtasks, propaganda span identification and propaganda technique classification. Our model utilizes recurrent Bi-LSTM layers with pre-trained word representations and also takes advantage of self-attention mechanism. Our model managed to achieve score 0.405 F1 for subtask 1 and 0.553 F1 for subtask 2 on test set resulting in 17th and 16th place in subtask 1 and subtask 2, respectively. 1 Introduction Nowadays, many publishers provide articles online to reach their audience faster. This shift from traditional journalism via printed press enabled more extensive misuse of objective news for own agenda using different propaganda techniques. Propagandistic articles try to convey their message using different techniques. Propaganda is the spread of purposefully shaped information in media to promote some sort of agenda (Da San Martino et al., 2019). This kind of information is produced by increasing number of biased news outlets in attempt to sway or manipulate public opinion. However, there is also a connected problem to propaganda, online spreading of fake news and misinformation in general (Simko et al., 2019). However, nowadays the widespread use of internet and access to multiple sources of information has caused traditional techniques of propaganda, like spreading of disinformation, to become far less effective. Along with this trend, propaganda techniques are naturally getting more complex to maximize their impact. Use of logical fallacies and appealing to the emotions of the audience is getting much more prominent because these techniques are significantly harder to spot at first glance and require much deeper analysis to get revealed. Research on detecting propaganda has focused primarily on news articles (Rashkin et al., 2017; Barr´on- Cedeno et al., 2019). However, analysis of articles produced by propagandistic news outlets indicates that these sources can often publish objective articles in attempt to increase their credibility (Horne et al., 2018). Recently the attention has moved towards fragment and sentence level propaganda detection. A new problem has been introduced – to detect the use of specific propaganda techniques in articles along with large corpora to assess this problem (Da San Martino et al., 2019). To tackle this problem many different neural model architectures were proposed, such as fine-tuned BERT pre-trained model (Yoosuf and Yang, 2019), ensemble of BERT, Bi-LSTM, and XGBoost (Al-Omari et al., 2019), or ensemble of logistic regression, CNN and BERT (Gupta et al., 2019). The main goal of this SemEval task was to develop tools …

Content truncated to 500 words