Reasoning About the Unsaid: Misinformation Detection with Omission-Aware Graph Inference

TL;DR

This work tackles misinformation detection by highlighting the underexplored dimension of omissions. It introduces OmiGraph, an omission-aware graph framework that constructs an environment from semantically related contextual news, reasons over intra- and inter-source relations with LLM-assisted omission intents, and performs omission-guided message passing to uncover hidden deceptive patterns. By fusing omission features with conventional signals, it achieves significant gains on English and Chinese benchmarks and sheds light on common omission types. The study also explores simulation-based omissions when external contexts are unavailable, underscoring the practical impact of reasoning about the unsaid for robust misinformation mitigation.

Abstract

This paper investigates the detection of misinformation, which deceives readers by explicitly fabricating misleading content or implicitly omitting important information necessary for informed judgment. While the former has been extensively studied, omission-based deception remains largely overlooked, even though it can subtly guide readers toward false conclusions under the illusion of completeness. To pioneer in this direction, this paper presents OmiGraph, the first omission-aware framework for misinformation detection. Specifically, OmiGraph constructs an omission-aware graph for the target news by utilizing a contextual environment that captures complementary perspectives of the same event, thereby surfacing potentially omitted contents. Based on this graph, omission-oriented relation modeling is then proposed to identify the internal contextual dependencies, as well as the dynamic omission intents, formulating a comprehensive omission relation representation. Finally, to extract omission patterns for detection, OmiGraph introduces omission-aware message-passing and aggregation that establishes holistic deception perception by integrating the omission contents and relations. Experiments show that, by considering the omission perspective, our approach attains remarkable performance, achieving average improvements of +5.4% F1 and +5.3% ACC on two large-scale benchmarks.

Figures (6)

• Figure 1: Concept comparison between existing methods and our OmiGraph. In contrast to solely focusing on the presented information, OmiGraph extracts more complete deception features via omission-aware modeling, capturing omitted key information such as the background and causal relation, and thus enhancing the existing detectors.
• Figure 2: Overview of OmiGraph. Given a news piece, OmiGraph constructs an omission-aware graph based on the contextual environment (a). Then, omission-oriented relation modeling reasons over the graph nodes, identifying intra-source contextual dependencies and inter-source omission intents (b). Finally, an omission-guided message passing mechanism extracts omission-oriented deception features (c) to enhance conventional misinformation detectors.
• Figure 3: Performance comparison of base models with and without OmiGraph. The better results in each group using the same base model are bolded. The $\pm$ denotes the standard deviation, * indicates 0.005 significance level from a paired t-test comparing OmiGraph with its base model.
• Figure 4: Performance comparison of OmiGraph and its variants using BERT as the base detector on the English (left) and Chinese (right) datasets.
• Figure 5: Omission type distribution standardized using Z-scores in English (left) and Chinese (right) dataset.