Truth and Trust: Fake News Detection via Biosignals

TL;DR

The study investigates whether non-invasive biosignals (EDA and PPG) reflect truth processing and belief when people encounter climate-related claims. It introduces two classification tasks—binary veracity and four-class joint belief-veracity—using a novel multimodal dataset (EDA/PPG, video, self-reports) and benchmarks four models (KNN, LightGBM, NN, CNN). Results show EDA generally outperforms PPG for veracity detection, with KNN providing the most robust performance; however, the joint belief-veracity task markedly reduces accuracy, underscoring the challenge of capturing nuanced cognitive states from biosignals alone. The work advocates multimodal, personalized approaches for misinformation detection and releases a dataset to spur further research into psychophysiological responses to misinformation.

Abstract

Understanding how individuals physiologically respond to false information is crucial for advancing misinformation detection systems. This study explores the potential of using physiological signals, specifically electrodermal activity (EDA) and photoplethysmography (PPG), to classify both the veracity of information and its interaction with user belief. In a controlled laboratory experiment, we collected EDA and PPG signals while participants evaluated the truthfulness of climate-related claims. Each trial was labeled based on the objective truth of the claim and the participant's belief, enabling two classification tasks: binary veracity detection and a novel four-class joint belief-veracity classification. We extracted handcrafted features from the raw signals and trained several machine learning models to benchmark the dataset. Our results show that EDA outperforms PPG, indicating its greater sensitivity to physiological responses related to truth perception. However, performance significantly drops in the joint belief-veracity classification task, highlighting the complexity of modeling the interaction between belief and truth. These findings suggest that while physiological signals can reflect basic truth perception, accurately modeling the intricate relationships between belief and veracity remains a significant challenge. This study emphasizes the importance of multimodal approaches that incorporate psychological, physiological, and cognitive factors to improve fake news detection systems. Our work provides a foundation for future research aimed at enhancing misinformation detection via addressing the complexities of human belief and truth processing.

Figures (3)

• Figure 1: Physiological foundations for detecting responses to misinformation. Our study investigates whether internal physiological signals reflect how people process (mis)information. (a) The autonomic nervous system governs involuntary responses, such as heart rate and sweat gland activity, that accompany cognitive and emotional states. (b) These responses are measured non-invasively via electrodermal activity (EDA) and photoplethysmography (PPG), which reflect arousal, stress, and attention. (c) During the experiment, participants wore an EmotiBit sensor that continuously recorded EDA and PPG while they evaluated the veracity of climate-related claims. By mapping these physiological signals to belief and truth judgments, our work explores how bodily responses can inform human-centered misinformation detection systems.
• Figure 2: Overview of the dataset construction process used for veracity and belief classification tasks. The dataset comprises three modalities: (i) screen recordings of participants evaluating climate-related claims (video), (ii) physiological signals: electrodermal activity (EDA) and photoplethysmography (PPG), recorded in CSV format, and (iii) participants’ self-reported belief responses (CSV). Timestamps of each claim presentation are extracted from the videos to isolate relevant time windows, which are then used to align the corresponding physiological data. Each trial is labeled with both a belief label (Believe or Not Believe) and a veracity label (True or False), enabling two downstream tasks: binary veracity classification and four-class joint belief-veracity classification. This multimodal, time-aligned dataset supports analysis of physiological responses to misinformation and the complex interplay between belief and truth.
• Figure 3: Distribution of labels in our dataset. (a) Distribution of claim veracity labels: True (T) claims slightly outnumber False (F) claims. (b) Distribution across four combined belief-veracity conditions: T-NB (True–Not Believe), T-B (True–Believe), F-NB (False–Not Believe), and F-B (False–Believe).