PERCEIV: A MULTIMODAL DATASET FOR ADAPTIVE INFORMATION VISUALIZATION FROM BEHAVIOURAL AND NEUROPHYSIOLOGICAL SIGNALS

Adaptive information visualization (InfoVis) interfaces require reliable indicators of users’ cognitive state, yet progress is limited by a lack of large, time-synchronized datasets combining behaviour with neurophysiology in controlled tasks. We present PERCEIV (PERCeption of visual Encodings in InfoVis), a large-scale multimodal dataset designed to support research on adaptive InfoVis through the integration of behavioural and neurophysiological signals. PERCEIV includes recordings from 120 participants performing InfoVis tasks under systematically varied difficulty and visualization conditions, while synchronized electroencephalography (EEG), eye tracking, and electrodermal activity (EDA) were acquired alongside behavioural responses. All modalities are time-aligned and accompanied by event markers, trial-wise annotations, and participant metadata. We release raw sensor streams together with processed, event-aligned derivatives and quality-control outputs to facilitate both method development and applied modelling. To lower the barrier to entry, we also provide baseline analyses and open-source scripts for preprocessing, feature extraction, and cognitive-load modelling across modalities (e.g., spectral EEG features, gaze dynamics, and EDA response metrics), as well as code for training and evaluating baseline machine-learning classifiers. Standardized data splits and comprehensive documentation enable reproducible benchmarking and direct comparison across studies. Initial baseline results show that multimodal models outperform single-modality baselines for discriminating task difficulty and workload-related conditions, with complementary contributions from gaze and EEG features. PERCEIV aims to accelerate research on multimodal sensing, adaptive interface policies, and the dynamics of cognitive state during information processing in visualization tasks.