RP-DNN: A Tweet Level Propagation Context Based Deep Neural Networks for Early Rumor Detection in Social Media

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6094–6105 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6094 RP-DNN: A Tweet level propagation context based deep neural networks for early rumor detection in Social Media Jie Gao, Sooji Han, Xingyi Song, Fabio Ciravegna Regent Court, 211 Portobello, Sheffield, UK, S1 4DP {j.gao, sooji.han, x.song, f.ciravegna}@sheffield.ac.uk Abstract Early rumor detection (ERD) on social media platform is very challenging when limited, incomplete and noisy information is available. Most of the existing methods have largely worked on event-level detection that requires the collection of posts relevant to a specific event and relied only on user-generated content. They are not suitable for detecting rumor sources in the very early stages, before an event unfolds and becomes widespread. In this paper, we address the task of ERD at the message level. We present a novel hybrid neural network architecture, which combines a task-specific character-based bidirectional language model and stacked Long Short-Term Memory (LSTM) networks to represent textual contents and social-temporal contexts of input source tweets, for modelling propagation patterns of rumors in the early stages of their development. We apply multi-layered attention models to jointly learn attentive context embeddings over multiple context inputs. Our experiments employ a stringent leave-one-out cross-validation (LOO-CV) evaluation set-up on seven publicly available real-life rumor event data sets. Our models achieve state-of-the-art(SoA) performance for detecting unseen rumors on large augmented data which covers more than 12 events and 2,967 rumors. An ablation study is conducted to understand the relative contribution of each component of our proposed model. Keywords: Early Rumor Detection, Social Media, Recurrent Neural Network, Attention Mechanism, Context Modeling 1. Introduction Research on social media rumors has become increasingly popular to understand the emergence and development of rumor events. An automatic and efficient approach for the early identification of rumors is vitally necessary in order to limit their spreading and minimize their effects. A typical rumor resolution process can include four sub- tasks: rumor detection, tracking, stance classification, and verification (Zubiaga et al., 2018). Rumor detection which aims to identify whether a claim is a rumor or non-rumor is a fundamental task for rumor resolution. Once a ru- mor is identified, it becomes possible to track its evolution over time, identify its sources, perform stance detection, and finally determine the its veracity (Zubiaga et al., 2018) (Kochkina et al., 2018). Recent research on online rumors has largely focused on the later stages of the process, that is, stance classification and verification. Although these are crucial for rumor resolution, they cannot be performed until rumors are identified. Several studies skip this preliminary task, either leaving the development of approaches for them for future work or assuming that rumors and their associ- ated posts are manual inputs. In this work, we highlight the importance of developing automated ERD systems for the success of the entire rumor resolution process. We propose a hybrid and context-aware deep neural net- work framework for tweet-level ERD, …

Content truncated to 500 words