Detecting Fake News Belief via Skin and Blood Flow Signals

TL;DR

The paper addresses how people physiologically respond to misinformation and repeated exposure, introducing a dataset linking EDA and PPG to belief and repetition judgments. Using an encoding–delay–test design with climate-related claims, they collect 672 trials from 28 participants wearing EmotiBit to capture EDA/PPG signals. They evaluate KNN, CNN, and LightGBM on EDA and PPG features, finding that EDA consistently provides stronger signals (best belief: 67.83% with KNN on EDA) and that four-class predictions are more challenging. The results suggest wearable biosignals can enable real-time, minimally invasive detection of belief and familiarity in misinformation, paving the way for trust-aware HCI and adaptive misinformation defenses.

Abstract

Misinformation poses significant risks to public opinion, health, and security. While most fake news detection methods rely on text analysis, little is known about how people physically respond to false information or repeated exposure to the same statements. This study investigates whether wearable sensors can detect belief in a statement or prior exposure to it. We conducted a controlled experiment where participants evaluated statements while wearing an EmotiBit sensor that measured their skin conductance (electrodermal activity, EDA) and peripheral blood flow (photoplethysmography, PPG). From 28 participants, we collected a dataset of 672 trials, each labeled with whether the participant believed the statement and whether they had seen it before. This dataset introduces a new resource for studying physiological responses to misinformation. Using machine learning models, including KNN, CNN, and LightGBM, we analyzed these physiological patterns. The best-performing model achieved 67.83\% accuracy, with skin conductance outperforming PPG. These findings demonstrate the potential of wearable sensors as a minimally intrusive tool for detecting belief and prior exposure, offering new directions for real-time misinformation detection and adaptive, user-aware systems.

Figures (6)

• Figure 1: Experimental setup for physiological signal collection using the EmotiBit sensor. (a) Close-up of the EmotiBit wearable sensor attached to the participant's forearm (b) A participant (c) Data acquisition workflow.
• Figure 2: Data annotation process. Raw EDA/PPG signals, screen recordings, and participant self-reports were synchronized based on timestamps to align with each claim. These synchronized intervals were then used to segment the EDA data and assign belief and repetition labels to each segment, creating a structured dataset at the trial level.
• Figure 3: Label % distribution across the dataset. (a) Distribution of the Belief and Repetition labels. The dataset has a slight skew to the Not Believed class, Repetition is nearly balanced. (b) Distribution across four combined conditions: Believed and Repeated (B-R), Believed and Not Repeated (B-NR), Not Believed and Repeated (NB-R), and Not Believed and Not Repeated (NB-NR).
• Figure 4: Visualizations of raw EDA and PPG signals for individual trials. Ttop row: EDA signals, Bottom row: corresponding PPG signals from the same trials. Each vertically aligned EDA–PPG pair comes from the same participant and corresponds to the same claim evaluation task. The $y$-axis indicates signal amplitude, and the $x$-axis represents time (in seconds).
• Figure 5: Data processing and classification pipeline.