DUCK: Rumour Detection on Social Media by Modelling User and Comment Propagation Networks

Abstract

Social media rumours, a form of misinformation, can mislead the public and cause significant economic and social disruption. Motivated by the observation that the user network -which captures who engages with a storyand the comment network -which captures how they react to it -provide complementary signals for rumour detection. In this paper, we propose DUCK (rumour detection with user and comment networks) for rumour detection on social media. We study how to leverage transformers and graph attention networks to jointly model the contents and the structure of social media conversations, as well as the network of users who engage in these conversations. Over four widely used benchmark rumour datasets in English and Chinese, we show that DUCK produces superior performance for detecting rumours, creating a new state-of-the-art.

Qualitative Analysis of a Graph Transformer Approach to Addressing Hate Speech: Adapting to Dynamically Changing Content

“…By taking a holistic view of conversations by encoding images, text, and discussion structure together, we hypothesize those hate speech detection methods would be able to avoid many false predictions, such as the ones incurred in Table 3. Furthermore, following Tian, Zhang, and Lau (2022), it would be possible to include user-level information into this graph representation.…”

“…This can result in predictions that focus more on the immediate discussion context rather than the larger global context. Other work has examined graph-based approaches for hate speech detection (Mishra et al 2019;Tian, Zhang, and Lau 2022); we confine our attention here to comparisons with the models already described above.…”

Heterogeneous graph neural networks with denoising for graph embeddings

“…5c). The directions of propagation are differentiated in some papers [1], [22], [35] to produce finer-grained propagation modelling, but the undirected graphs are the main stream in static graph-based methods.…”

“…DUCK [35] claim, comments, retweets, users responsive propagation GAT Consider temporal and structural information of comments and retweets in propagation. UniPF [36] claims, comments, topics cluster connection, responsive propagation Adapt-GCN [36] Connect claims under the same topic.…”

“…Considering that the patterns can contain noise and be modified by fake news spreaders, edge enhancement, adversarial learning, and contrastive learning strategies are often used [26]- [30]. There is also a trend to combine more information with the propagation patterns, such as temporal information [37]- [40], [42], [43] and user information [32], [33], [35].…”