Widespread Chinese social media applications such as Weibo are widely known for monitoring and deleting posts to conform to Chinese government requirements. In this paper, we focus on analyzing a dataset of censored and uncensored posts in Weibo. Despite previous work that only considers text content of posts, we take a multi-modal approach that takes into account both text and image content. We categorize this dataset into 14 categories that have the potential to be censored on Weibo, and seek to quantify censorship by topic. Specifically, we investigate how different factors interact to affect censorship. We also investigate how consistently and how quickly different topics are censored. To this end, we have assembled an image dataset with 18,966 images, as well as a text dataset with 994 posts from 14 categories. We then utilized deep learning, CNN localization, and NLP techniques to analyze the target dataset and extract categories, for further analysis to better understand censorship mechanisms in Weibo. We found that sentiment is the only indicator of censorship that is consistent across the variety of topics we identified. Our finding matches with recently leaked logs from Sina Weibo. We also discovered that most categories like those related to anti-government actions (e.g. protest) or categories related to politicians (e.g. Xi Jinping) are often censored, whereas some categories such as crisis-related categories (e.g. rainstorm) are less frequently censored. We also found that censored posts across all categories are deleted in three hours on average.

Abusive text is a serious problem in social media and causes many issues among users as the number of users and the content volume increase. There are several attempts for detecting or preventing abusive text effectively. One simple yet effective approach is to use an abusive lexicon and determine the existence of an abusive word in text. This approach works well even when an abusive word is obfuscated. On the other hand, it is still a challenging problem to determine abusiveness in a text having no explicit abusive words. Especially, it is hard to identify sarcasm or offensiveness in context without any abusive words. We tackle this problem using an ensemble deep learning model. Our model consists of two parts of extracting local features and global features, which are crucial for identifying implicit abusiveness in context level. We evaluate our model using three benchmark data. Our model outperforms all the previous models for detecting abusiveness in a text data without abusive words. Furthermore, we combine our model and an abusive lexicon method. The experimental results show that our model has at least 4% better performance compared with the previous approaches for identifying text abusiveness in case of with/without abusive words.

Social media has reportedly been (ab)used by Russian troll farms to promote political agendas. Specifically, state-affiliated actors disguise themselves as native citizens of the United States to promote discord and promote their political motives. Therefore, developing methods to automatically detect Russian trolls can ensure fair elections and possibly reduce political extremism by stopping trolls that produce discord. While data exists for some troll organizations (e.g., Internet Research Agency), it is challenging to collect ground-truth accounts for new troll farms in a timely fashion. In this paper, we study the impact the number of labeled troll accounts has on detection performance. We analyze the use of self-supervision with less than 100 troll accounts as training data. We improve classification performance by nearly 4% F1. Furthermore, in combination with self-supervision, we also explore novel features for troll detection grounded in stylometry. Intuitively, we assume that the writing style is consistent across troll accounts because a single troll organization employee may control multiple user accounts. Overall, we improve on models based on words features by ~9% F1.

The blurry line between nefarious fake news and protected-speech satire has been a notorious struggle for social media platforms. Further to the efforts of reducing exposure to misinformation on social media, purveyors of fake news have begun to masquerade as satire sites to avoid being demoted. In this work, we address the challenge of automatically classifying fake news versus satire. Previous work have studied whether fake news and satire can be distinguished based on language differences. Contrary to fake news, satire stories are usually humorous and carry some political or social message. We hypothesize that these nuances could be identified using semantic and linguistic cues. Consequently, we train a machine learning method using semantic representation, with a state-of-the-art contextual language model, and with linguistic features based on textual coherence metrics. Empirical evaluation attests to the merits of our approach compared to the language-based baseline and sheds light on the nuances between fake news and satire. As avenues for future work, we consider studying additional linguistic features related to the humor aspect, and enriching the data with current news events, to help identify a political or social message.

Calls to action on social media are known to be effective means of mobilization in social movements, and a frequent target of censorship. We investigate the possibility of their automatic detection and their potential for predicting real-world protest events, on historical data of Bolotnaya protests in Russia (2011-2013). We find that political calls to action can be annotated and detected with relatively high accuracy, and that in our sample their volume has a moderate positive correlation with rally attendance.

Digital media enables not only fast sharing of information, but also disinformation. One prominent case of an event leading to circulation of disinformation on social media is the MH17 plane crash. Studies analysing the spread of information about this event on Twitter have focused on small, manually annotated datasets, or used proxys for data annotation. In this work, we examine to what extent text classifiers can be used to label data for subsequent content analysis, in particular we focus on predicting pro-Russian and pro-Ukrainian Twitter content related to the MH17 plane crash. Even though we find that a neural classifier improves over a hashtag based baseline, labeling pro-Russian and pro-Ukrainian content with high precision remains a challenging problem. We provide an error analysis underlining the difficulty of the task and identify factors that might help improve classification in future work. Finally, we show how the classifier can facilitate the annotation task for human annotators.

Considering diverse aspects of an argumentative issue is an essential step for mitigating a biased opinion and making reasonable decisions. A related generation model can produce flexible results that cover a wide range of topics, compared to the retrieval-based method that may show unstable performance for unseen data. In this paper, we study the problem of generating sentential arguments from multiple perspectives, and propose a neural method to address this problem. Our model, ArgDiver (Argument generation model from diverse perspectives), in a way a conversational system, successfully generates high-quality sentential arguments. At the same time, the automatically generated arguments by our model show a higher diversity than those generated by any other baseline models. We believe that our work provides evidence for the potential of a good generation model in providing diverse perspectives on a controversial topic.

Social media platforms have been used for information and news gathering, and they are very valuable in many applications. However, they also lead to the spreading of rumors and fake news. Many efforts have been taken to detect and debunk rumors on social media by analyzing their content and social context using machine learning techniques. This paper gives an overview of the recent studies in the rumor detection field. It provides a comprehensive list of datasets used for rumor detection, and reviews the important studies based on what types of information they exploit and the approaches they take. And more importantly, we also present several new directions for future research.

Many discussions on online platforms suffer from users offending others by using abusive terminology, threatening each other, or being sarcastic. Since an automatic detection of abusive language can support human moderators of online discussion platforms, detecting abusiveness has recently received increased attention. However, the existing approaches simply train one classifier for the whole variety of abusiveness. In contrast, our approach is to distinguish explicitly abusive cases from the more “shadowed” ones. By dynamically extending a lexicon of abusive terms (e.g., including new obfuscations of abusive terms), our approach can support a moderator with explicit unraveled explanations for why something was flagged as abusive: due to known explicitly abusive terms, due to newly detected (obfuscated) terms, or due to shadowed cases.

The goal of fine-grained propaganda detection is to determine whether a given sentence uses propaganda techniques (sentence-level) or to recognize which techniques are used (fragment-level). This paper presents the sys- tem of our participation in the sentence-level subtask of the propaganda detection shared task. In order to better utilize the document information, we construct context-dependent input pairs (sentence-title pair and sentence- context pair) to fine-tune the pretrained BERT, and we also use the undersampling method to tackle the problem of imbalanced data.

This paper presents the winning solution of the Fragment Level Classification (FLC) task in the Fine Grained Propaganda Detection competition at the NLP4IF’19 workshop. The goal of the FLC task is to detect and classify textual segments that correspond to one of the 18 given propaganda techniques in a news articles dataset. The main idea of our solution is to perform word-level classification using fine-tuned BERT, a popular pre-trained language model. Besides presenting the model and its evaluation results, we also investigate the attention heads in the model, which provide insights into what the model learns, as well as aspects for potential improvements.

This paper describes our system (MIC-CIS) details and results of participation in the fine grained propaganda detection shared task 2019. To address the tasks of sentence (SLC) and fragment level (FLC) propaganda detection, we explore different neural architectures (e.g., CNN, LSTM-CRF and BERT) and extract linguistic (e.g., part-of-speech, named entity, readability, sentiment, emotion, etc.), layout and topical features. Specifically, we have designed multi-granularity and multi-tasking neural architectures to jointly perform both the sentence and fragment level propaganda detection. Additionally, we investigate different ensemble schemes such as majority-voting, relax-voting, etc. to boost overall system performance. Compared to the other participating systems, our submissions are ranked 3rd and 4th in FLC and SLC tasks, respectively.

This paper presents the CUNLP submission for the NLP4IF 2019 shared-task on Fine-Grained Propaganda Detection. Our system finished 5th out of 26 teams on the sentence-level classification task and 5th out of 11 teams on the fragment-level classification task based on our scores on the blind test set. We present our models, a discussion of our ablation studies and experiments, and an analysis of our performance on all eighteen propaganda techniques present in the corpus of the shared task.

On the NLP4IF 2019 sentence level propaganda classification task, we used a BERT language model that was pre-trained on Wikipedia and BookCorpus as team ltuorp ranking #1 of 26. It uses deep learning in the form of an attention transformer. We substituted the final layer of the neural network to a linear real valued output neuron from a layer of softmaxes. The backpropagation trained the entire neural network and not just the last layer. Training took 3 epochs and on our computation resources this took approximately one day. The pre-trained model consisted of uncased words and there were 12-layers, 768-hidden neurons with 12-heads for a total of 110 million parameters. The articles used in the training data promote divisive language similar to state-actor-funded influence operations on social media. Twitter shows state-sponsored examples designed to maximize division occurring across political lines, ranging from “Obama calls me a clinger, Hillary calls me deplorable, ... and Trump calls me an American” oriented to the political right, to Russian propaganda featuring “Black Lives Matter” material with suggestions of institutional racism in US police forces oriented to the political left. We hope that raising awareness through our work will reduce the polarizing dialogue for the betterment of nations.

Bias is ubiquitous in most online sources of natural language, from news media to social networks. Given the steady shift in news consumption behavior from traditional outlets to online sources, the automatic detection of propaganda, in which information is shaped to purposefully foster a predetermined agenda, is an increasingly crucial task. To this goal, we explore the task of sentence-level propaganda detection, and experiment with both handcrafted features and learned dense semantic representations. We also experiment with random undersampling of the majority class (non-propaganda) to curb the influence of class distribution on the system’s performance, leading to marked improvements on the minority class (propaganda). Our best performing system uses pre-trained ELMo word embeddings, followed by a bidirectional LSTM and an attention layer. We have submitted a 5-model ensemble of our best performing system to the NLP4IF shared task on sentence-level propaganda detection (team LIACC), achieving rank 10 among 25 participants, with 59.5 F1-score.

The internet and the high use of social media have enabled the modern-day journalism to publish, share and spread news that is difficult to distinguish if it is true or fake. Defining “fake news” is not well established yet, however, it can be categorized under several labels: false, biased, or framed to mislead the readers that are characterized as propaganda. Digital content production technologies with logical fallacies and emotional language can be used as propaganda techniques to gain more readers or mislead the audience. Recently, several researchers have proposed deep learning (DL) models to address this issue. This research paper provides an ensemble deep learning model using BiLSTM, XGBoost, and BERT to detect propaganda. The proposed model has been applied on the dataset provided by the challenge NLP4IF 2019, Task 1 Sentence Level Classification (SLC) and it shows a significant performance over the baseline model.

Various propaganda techniques are used to manipulate peoples perspectives in order to foster a predetermined agenda such as by the use of logical fallacies or appealing to the emotions of the audience. In this paper, we develop a Logistic Regression-based tool that automatically classifies whether a sentence is propagandistic or not. We utilize features like TF-IDF, BERT vector, sentence length, readability grade level, emotion feature, LIWC feature and emphatic content feature to help us differentiate these two categories. The linguistic and semantic features combination results in 66.16% of F1 score, which outperforms the baseline hugely.

The automatic identification of propaganda has gained significance in recent years due to technological and social changes in the way news is generated and consumed. That this task can be addressed effectively using BERT, a powerful new architecture which can be fine-tuned for text classification tasks, is not surprising. However, propaganda detection, like other tasks that deal with news documents and other forms of decontextualized social communication (e.g. sentiment analysis), inherently deals with data whose categories are simultaneously imbalanced and dissimilar. We show that BERT, while capable of handling imbalanced classes with no additional data augmentation, does not generalise well when the training and test data are sufficiently dissimilar (as is often the case with news sources, whose topics evolve over time). We show how to address this problem by providing a statistical measure of similarity between datasets and a method of incorporating cost-weighting into BERT when the training and test sets are dissimilar. We test these methods on the Propaganda Techniques Corpus (PTC) and achieve the second highest score on sentence-level propaganda classification.

In this paper, we describe our system used in the shared task for fine-grained propaganda analysis at sentence level. Despite the challenging nature of the task, our pretrained BERT model (team YMJA) fine tuned on the training dataset provided by the shared task scored 0.62 F1 on the test set and ranked third among 25 teams who participated in the contest. We present a set of illustrative experiments to better understand the performance of our BERT model on this shared task. Further, we explore beyond the given dataset for false-positive cases that likely to be produced by our system. We show that despite the high performance on the given testset, our system may have the tendency of classifying opinion pieces as propaganda and cannot distinguish quotations of propaganda speech from actual usage of propaganda techniques.

In this paper, we describe our team’s effort on the fine-grained propaganda detection on sentence level classification (SLC) task of NLP4IF 2019 workshop co-located with the EMNLP-IJCNLP 2019 conference. Our top performing system results come from applying ensemble average on three pretrained models to make their predictions. The first two models use the uncased and cased versions of Bidirectional Encoder Representations from Transformers (BERT) (Devlin et al., 2018) while the third model uses Universal Sentence Encoder (USE) (Cer et al. 2018). Out of 26 participating teams, our system is ranked in the first place with 68.8312 F1-score on the development dataset and in the sixth place with 61.3870 F1-score on the testing dataset.

In this paper, we describe our approach and system description for NLP4IF 2019 Workshop: Shared Task on Fine-Grained Propaganda Detection. Given a sentence from a news article, the task is to detect whether the sentence contains a propagandistic agenda or not. The main contribution of our work is to evaluate the effectiveness of various transfer learning approaches like ELMo, BERT, and RoBERTa for propaganda detection. We show the use of Document Embeddings on the top of Stacked Embeddings combined with LSTM for identification of propagandistic context in the sentence. We further provide analysis of these models to show the effect of oversampling on the provided dataset. In the final test-set evaluation, our system ranked 21st with F1-score of 0.43 in the SLC Task.

In recent years, the need for communication increased in online social media. Propaganda is a mechanism which was used throughout history to influence public opinion and it is gaining a new dimension with the rising interest of online social media. This paper presents our submission to NLP4IF-2019 Shared Task SLC: Sentence-level Propaganda Detection in news articles. The challenge of this task is to build a robust binary classifier able to provide corresponding propaganda labels, propaganda or non-propaganda. Our model relies on a unified neural network, which consists of several deep leaning modules, namely BERT, BiLSTM and Capsule, to solve the sentencelevel propaganda classification problem. In addition, we take a pre-training approach on a somewhat similar task (i.e., emotion classification) improving results against the cold-start model. Among the 26 participant teams in the NLP4IF-2019 Task SLC, our solution ranked 12th with an F1-score 0.5868 on the official test data. Our proposed solution indicates promising results since our system significantly exceeds the baseline approach of the organizers by 0.1521 and is slightly lower than the winning system by 0.0454.

This paper presents a method of detecting fine-grained categories of propaganda in text. Given a sentence, our method aims to identify a span of words and predict the type of propaganda used. To detect propaganda, we explore a method for extracting features of propaganda from contextualized embeddings without fine-tuning the large parameters of the base model. We show that by generating synthetic embeddings we can train a linear function with ReLU activation to extract useful labeled embeddings from an embedding space generated by a general-purpose language model. We also introduce an inference technique to detect continuous spans in sequences of propaganda tokens in sentences. A result of the ensemble model is submitted to the first shared task in fine-grained propaganda detection at NLP4IF as Team Stalin. In this paper, we provide additional analysis regarding our method of detecting spans of propaganda with synthetically generated representations.

We present the shared task on Fine-Grained Propaganda Detection, which was organized as part of the NLP4IF workshop at EMNLP-IJCNLP 2019. There were two subtasks. FLC is a fragment-level task that asks for the identification of propagandist text fragments in a news article and also for the prediction of the specific propaganda technique used in each such fragment (18-way classification task). SLC is a sentence-level binary classification task asking to detect the sentences that contain propaganda. A total of 12 teams submitted systems for the FLC task, 25 teams did so for the SLC task, and 14 teams eventually submitted a system description paper. For both subtasks, most systems managed to beat the baseline by a sizable margin. The leaderboard and the data from the competition are available at http://propaganda.qcri.org/nlp4if-shared-task/.