Research Foundations

Built on four decades of
established science.

EMPATHIQ synthesises theoretical frameworks from behavioural economics, neuroscience, and affective computing with recent advances in wearable sensing and machine learning. The invention addresses a gap identified consistently across the literature: every major institution has solved one piece of the puzzle in isolation.

This page is openly published without registration. It covers the published academic evidence base, the theoretical foundations, the research gap analysis, and the research agenda. It does not disclose the patented system architecture or implementation details.

Theoretical Foundations

Four frameworks. One integrated system.

These are not peripheral references — they are the direct theoretical architecture of the invention. Each framework maps to a specific technical component.

S-O-R
Stimulus-Organism-Response Model
Mehrabian & Russell, 1974 · Consumer behaviour
Environmental and internal stimuli influence the organism's internal affective state, which drives response behaviour. The theoretical basis for the entire sensing → ESV → intervention architecture of EMPATHIQ. Extensively validated in consumer research through 2025.
→ Directly supports Layers 1–4
PAD
Pleasure-Arousal-Dominance Model
Mehrabian & Russell, 1974 · Affective computing
The three-dimensional model of affect used as the structure of the Emotional State Vector (ESV). Valence (pleasure), Arousal, and Dominance provide the dimensional space in which emotional states are measured, compared, and tracked over time.
→ Direct structure of the ESV output
SMH
Somatic Marker Hypothesis
Damasio, 1994 · Cognitive neuroscience
Emotion-based physiological signals arising from the body are integrated in the ventromedial prefrontal cortex to regulate decision-making under uncertainty. Provides the neurobiological basis for using HRV and EDA biosignals as proxies for decision quality risk.
→ Justifies biosignal sensing for RDRS
NIM
Neurovisceral Integration Model
Thayer & Lane, 2009 · Psychophysiology
Vagally-mediated heart rate variability (HRV) is a biomarker of prefrontal inhibitory control capacity — the neural mechanism that enables deliberative decision-making. Directly supports HRV as the primary physiological input to the RDRS computation and PEB.
→ Justifies HRV as primary RDRS input
Evidence Base

Nine research foundations. All peer-reviewed.

EMPATHIQ's research brief documents the state of the field across nine major research areas. The consistent gap: every institution has solved one piece. No one has combined them.

META
ANALYSIS
Journal of Academy of Marketing Science · 231 samples · 75,000+ consumers
Impulse Buying Meta-Analysis — Population-Scale Evidence
Key finding: High-arousal emotional states and hedonic motives are primary triggers of impulsive purchasing behaviour. Self-control and mood state confirmed as the key mediating mechanisms — emotion is the lever. The strongest meta-analytic evidence to date that the problem EMPATHIQ addresses is real and large-scale.
Gap: Confirms the phenomenon at population level — provides no individual-level early detection mechanism, no personalised risk model, and no real-time intervention capability. Entirely retrospective.
Core Problem Evidence
META
REGRESSION
PMC 2023 · 51 studies · 14,957 participants · Systematic review
Emotion-Related Impulsivity Meta-Regression
Key finding: A consistent positive relationship between emotion-related impulsivity and risky decision-making, generalising across age, gender, positive and negative emotional valence, and clinical vs. non-clinical populations. Physiological arousal manipulation identified as a near-significant moderator.
Gap: Identifies the correlation. Does not propose a detection-and-intervention architecture. Explicitly recommends future research identify specific types of arousal inductions that best capture emotion-related impulsivity — the empirical programme EMPATHIQ's DEH is designed to fill.
RDRS Foundation
AHER
2024
Carnegie Mellon · Imperial College · Tsinghua · Neural Computing & Applications
Automated Human Emotion Recognition — Comprehensive Review
Key finding: ML systems using facial, speech, EEG, ECG, and GSR achieve classification accuracies exceeding 85% across multiple emotional dimensions. Emotion is confirmed as a critical driver of decision-making, planning, and reasoning.
Gap: Research stops at recognition. Does not connect emotional states to downstream behaviour change, personal history, or proactive intervention.
Sensing Layer
EEG
2025
Multi-institution PRISMA systematic review · 64 studies · PubMed, Scopus, Web of Science
EEG-Based Emotion Recognition & Neurofeedback
Key finding: Multimodal approaches combining EEG with physiological signals surpass 90% classification accuracy in controlled environments. Wearable EEG devices (MUSE 2) have opened real-world continuous emotion monitoring pathways outside the lab.
Gap: Confined to clinical or research settings. No longitudinal personal profile, no decision-event linkage, no conversational AI layer.
EEG Modality
WEAR
2024
MIT Media Lab · ETH Zurich Wearable Computing · Stanford HCI Group · PMC Scoping Review
Wearable Biosignal Stress Detection
Key finding: Samsung Galaxy Watch, Empatica E4, Polar H10 can continuously capture HRV, EDA, BVP, and accelerometer data. These signals correlate strongly with stress and emotional arousal. Battery life now extends to 16 days — continuous daily monitoring is commercially feasible.
Gap: Data collected but not acted upon in real time. No feedback loop to the user at the moment of a consequential decision.
Hardware Readiness
EMO
DEC
Hong Kong University of Science & Technology · University of Edinburgh · ScienceDirect 2025
Emotion Dynamics & Decision Making
Key finding: Online decision-making is deeply intertwined with emotional state fluctuations. Generative AI (GPT-4, RoBERTa) can deliver personalised emotional regulation interventions that measurably improve decision quality.
Gap: Operates only in online/information-retrieval contexts. No persistent personal history, no physiological data layer, no agentic proactive capability.
Intervention Layer
AFF
2025
University of Amsterdam · King's College London · PMC Narrative Review 1997–2024
Affective Computing for Mental Health
Key finding: Affective computing reduces manual tracking burden and enables continuous emotional data collection. Real-time analysis provides instantaneous feedback on mood fluctuations. Platforms combining passive sensing with mood self-reports demonstrate improved self-awareness.
Gap: Reactive and clinical. Lacks the anticipatory intelligence needed to predict and prevent rash decisions before they happen.
Wellbeing Layer
EMA
METH
Vrije Universiteit Amsterdam / Amsterdam UMC · PRISMA systematic review · 53 smartphone EMA studies
Ecological Momentary Assessment Methodology
Key finding: EMA — the systematic repeated measurement of psychological states in daily life using smartphones and wearables — is the gold-standard methodology for capturing real-world emotional fluctuations. Average study: 12.8 days, 2–12 assessments per day. Mean compliance: 71.6% — sufficient for longitudinal modelling.
Gap: Existing EMA research does not link momentary emotional assessments to decision events and their outcomes. EMPATHIQ's DEH is the proposed integration.
DEH Methodology
IEEE
2023
Pennsylvania State University · Tsinghua University BNRist · IEEE 2023
Emotional AI in Human-Computer Interaction
Key finding: Artificial emotional intelligence remains in its infancy for real-world continuous personalised application. Authors explicitly call for culturally adaptive, privacy-preserving, longitudinal emotion systems that connect emotional understanding to meaningful life outcomes.
Gap: No implementation framework for decision support or personal AI guardian. The call for cultural adaptivity is important: ~73% of current literature is from China, US, and EU — a WEIRD skew requiring cross-cultural validation before global deployment.
Cultural Validity
Gap Analysis

What nobody has built yet.

The researcher skill gap analysis framework (Importance × Tractability × Novelty, max 125) applied to EMPATHIQ's core gaps.

Integration Gap · Score: 110/125
DEH — Decision-Emotion Linkage Dataset
No dataset exists linking continuous physiological emotional data to real-world personal decision events and outcomes over time. Field A (emotion recognition) and Field B (decision research) have never cross-referenced at the individual level.
Importance
5/5
Tractability
4.4/5
Novelty
5/5
Knowledge Gap · Score: 100/125
RDRS — Personalised Decision Risk Classifier
No individual-level predictive model has been built that uses a person's own emotional history to predict decision quality risk in real time. Population-level models exist; personal models do not.
Importance
5/5
Tractability
4/5
Novelty
5/5
Population Gap · Score: 88/125
Cross-Cultural Validity
~73% of the emotion recognition and impulse-decision literature comes from China, US, and EU. The PAD model, ESV thresholds, and decision regret definitions require validation across non-WEIRD populations before global deployment.
Importance
4.4/5
Tractability
4/5
Novelty
5/5
Research Agenda

Six research programmes ready to begin.

01
Priority: Critical
Personalised Decision-Emotion Linkage Dataset Using EMA
Design and execute the world's first longitudinal EMA study linking continuous physiological emotional data to real-world decision events and their outcomes. Minimum 6 months, 200+ participants, continuous wearable biosignals, decision event EMA prompts, 24–72h outcome retrospectives.
Methodology: EMA · Wearable biosignals · Longitudinal · N=250 target (expecting 180+ complete)
02
Priority: Critical
Rash Decision Classifier — Operationalisation and Training
Formally operationalise "rash decision" (immediate regret? financial loss? reversal behaviour?) and build a labelled training corpus from the EMA dataset. Develop the personalised LSTM architecture. Validate on held-out personal data per participant. Establish minimum DEH records required for reliable RDRS performance.
Methodology: LSTM · Personalised ML · Held-out validation · Weighed BCE loss
03
Priority: High
Personalised Baseline Adaptation Algorithms
Develop and validate algorithms for rapid individual baseline construction and continuous adaptation as users change over time — seasonal mood shifts, life events, ageing. Evaluate forgetting factor (α) values across demographic profiles. Validate federated learning gradient update protocol against individual data reconstruction attacks.
Methodology: Online learning · Federated learning · Differential privacy · Longitudinal stability analysis
04
Priority: High
Intervention Efficacy Studies — Tiered RCT Design
Pre-registered randomised controlled trials comparing no intervention (control), soft nudge, Pattern Mirror, pause protocol, and wellbeing bridge — across personality types and cultural contexts. Primary outcome: decision regret rate. Secondary: financial outcomes, wellbeing measures, user acceptance. Report Cohen's d and confidence intervals.
Methodology: RCT · Pre-registration (OSF) · Personality × Culture factorial design · N=400+ recommended
05
Priority: High
Privacy-Preserving Architecture Validation
Audit the federated learning gradient update protocol to confirm that (ε,δ)-differential privacy guarantees hold under adversarial conditions. Validate that no individual ESV, DEH, or decision history can be reconstructed from transmitted gradient updates. Satisfy EU AI Act high-risk obligations and OAIC Privacy Impact Assessment requirements before Phase 1 deployment.
Methodology: Differential privacy formal proof · Red team testing · OAIC PIA framework
06
Priority: Required Before Global Launch
Cross-Cultural Validity
Validate whether the PAD model applies equivalently across non-WEIRD populations. Assess whether physiological signals associated with high-arousal states have consistent expression across cultural contexts. Determine whether "rash decision" has equivalent meaning and operationalisation across cultural groups with different risk tolerances and decision-making norms.
Methodology: Cross-cultural EMA study · Measurement invariance testing · Conceptual equivalence assessment · Minimum 6 countries
Research Collaboration

Looking for the right research partners.

ARC Linkage / Industry PhD
Australian Universities
The EMPATHIQ research programme is structured for an ARC Linkage Project or Industry PhD. We are looking for supervisors with expertise in affective computing, HCI, decision neuroscience, or consumer psychology. The research agenda is well-specified, the patent is filed, and the commercial case is strong — an ideal Linkage project profile.
  • ▸ARC Linkage Grant co-application
  • ▸Industry PhD supervision in affective computing or HCI
  • ▸Research collaboration with CSIRO Data61
  • ▸Ethics committee application support for EMA studies
Enquire →
International Collaboration
Global Research Institutions
The cross-cultural validity agenda (Research Item 6) specifically requires non-Australian research partners. We are seeking institutions in Asia, Europe, the Middle East, and Latin America to co-design and co-execute cross-cultural EMA studies. Joint publication opportunities are available across all six research agenda items.
  • ▸Cross-cultural EMA study co-design
  • ▸Joint publication on DEH methodology
  • ▸Intervention efficacy RCT partnership
  • ▸RDRS validity and performance benchmarking
Enquire →