Mapping (Dis-)Information Flow about the MH17 Plane Crash
Proceedings of the Second Workshop on Natural Language Processing for Internet Freedom: Censorship, Disinformation, and Propaganda
ii Preface Welcome to the second edition of the Workshop on Natural Language Processing for Internet Freedom (NLP4IF 2019). This year, we focused on censorship, disinformation, and propaganda. We further featured a shared task on the identification of propaganda in news articles. The task included two subtasks with different levels of complexity. Given a news article, the FLC subtask (fragment-level classification) asked for the identification of the propagandistic text fragments and also for
... e prediction of the specific propaganda technique used in this fragment (18-way classification task). The SLC subtask (sentence-level classification) is a binary classification task, which asked to detect the sentences that contain propaganda. A total of 39 teams submitted runs; 21 teams participated in the FLC subtask and 35 teams took part in the SLC subtask. Fourteen participants submitted a system description paper, which include models based on a wide range of learning models (e.g., neural networks, logistic regression) and representations (e.g., manually-engineered features, distributional representations). We accepted a total of 24 papers: 10 for the regular track and 14 for the shared task. We are excited that the workshop includes a diverse set of topics: rumor and trolls detection, censorship and controversy, fake news vs. satire, uncovering propaganda and abusive language identification. We are also thrilled to be able to bring an invited speaker, Elissa Redmiles from Princeton University and Microsoft Research, with a talk on measuring human perception to defend democracy, exploring a specific attack on the freedom of U.S. elections -the IRA Facebook advertisements, which successfully influenced people and avoided detection -and a defense against propaganda, which uses human perceptions to defend against the very propaganda that aims to influence those perceptions. Last but not least, we would like to thank the program committee and the shared task participants for their help with reviewing the papers, and with advertising the workshop.