Datasets:

wincode
/

HESA-v2

language:
  - en
license: mit
task_categories:
  - text-classification
  - text-generation
  - feature-extraction
task_ids:
  - sentiment-analysis
  - intent-classification
pretty_name: HESA-v1 Human Emotion and System Adaptation Dataset
size_categories:
  - 1K<n<10K
tags:
  - emotion-ai
  - adaptive-ai
  - human-centered-ai
  - cognitive-ai
  - ai-operating-system
  - synthetic-dataset
  - emotion-detection
  - productivity-ai
  - agi
  - system-adaptation
dataset_info:
  features:
    - name: sample_id
      dtype: string
    - name: timestamp
      dtype: string
    - name: user_profile
      dtype: string
    - name: environment
      dtype: string
    - name: system_state
      dtype: string
    - name: human_behavior
      dtype: string
    - name: conversation_context
      dtype: string
    - name: ai_system_response
      dtype: string
    - name: long_term_memory
      dtype: string
  splits:
    - name: train
      num_examples: 400
    - name: validation
      num_examples: 60
    - name: test
      num_examples: 40
  download_size: 1904937
  dataset_size: 1904937
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train.json
      - split: validation
        path: data/validation.json
      - split: test
        path: data/test.json

🧠 HESA-v1

Human Emotion + System Adaptation Dataset

A synthetic, research-grade dataset for training emotion-aware AI systems, adaptive operating systems, and human-centered computing platforms.

📖 Dataset Description

HESA-v1 is a large-scale synthetic dataset designed to simulate the complex relationship between human emotional states, computer usage behavior, environmental context, and intelligent AI adaptive responses.

This dataset was engineered for training next-generation:

🤖 Emotion-aware Operating Systems
🧩 Adaptive AGI Systems
📚 Intelligent Tutoring Systems
💬 Emotionally Intelligent Assistants
🧠 Cognitive AI Systems
🖥️ Human-Centered Computing Platforms

Each sample captures a complete human-computer interaction snapshot — combining psychological state, environmental conditions, system telemetry, behavioral signals, and AI adaptive responses into one richly structured JSON record.

🗂️ Dataset Structure

Top-level Fields

{
  "sample_id": "HESA-00001",
  "timestamp": "2026-03-14T22:47:00",
  "user_profile": { ... },
  "environment": { ... },
  "system_state": { ... },
  "human_behavior": { ... },
  "conversation_context": { ... },
  "ai_system_response": { ... },
  "long_term_memory": { ... }
}

👤 `user_profile`

Field	Type	Description
`age`	int	User age (16–58)
`profession`	string	Job/role (30 unique professions)
`country`	string	Country of origin (25 countries)
`personality_type`	string	MBTI type (all 16 types)
`experience_level`	string	Beginner → Veteran
`sleep_hours`	float	Hours slept last night (3.0–10.0)
`stress_level`	string	Very Low → Critical
`mental_focus`	string	Scattered → Hyperfocused
`social_energy`	string	Depleted → Highly Social

🌍 `environment`

Field	Type	Description
`location_type`	string	Home Office, Library, Coffee Shop, etc.
`noise_level`	string	Silent → Chaotic
`lighting`	string	Pitch Dark → Fluorescent Harsh
`temperature`	string	Cold (15°C) → Hot (31°C)
`time_of_day`	string	9 slots from Early Morning to Deep Night
`weather`	string	Clear Sunny, Rainy, Stormy, Snowing, etc.

💻 `system_state`

Field	Type	Description
`cpu_usage_percent`	int	CPU load (8–97%)
`ram_usage_percent`	int	RAM utilization (25–92%)
`battery_level_percent`	int	Battery charge (4–100%)
`network_quality`	string	Offline → Gigabit Ethernet
`open_applications`	list	2–9 apps from 34 real applications
`active_application`	string	Currently focused app
`typing_speed_wpm`	int	WPM — derived from stress + sleep + focus
`mouse_movement_pattern`	string	Behavioral mouse pattern
`scroll_behavior`	string	Reading/skimming/jumping pattern
`error_frequency_rate`	float	Error rate 0.0–1.0 (derived)
`multitasking_level`	string	None → Extreme

🧑‍💻 `human_behavior`

Field	Type	Description
`emotion`	string	28 emotion categories
`emotion_intensity`	string	Subtle → Overwhelming
`facial_expression`	string	Observable facial cues
`voice_tone`	string	Vocal behavioral signal
`typing_pattern`	string	Keystroke behavioral pattern
`interaction_style`	string	How user engages with system
`frustration_signals`	list	Observable frustration indicators
`focus_duration_minutes`	int	Minutes in current focus block
`energy_level`	string	Depleted → Peak
`motivation_level`	string	None → Intrinsic Drive
`cognitive_load`	string	Minimal → Overloaded
`learning_state`	string	Not Learning → Breakthrough Moment

💬 `conversation_context`

Field	Type	Description
`user_message`	string	Realistic user query/request
`conversation_goal`	string	Intent behind the interaction
`topic`	string	Task or subject being worked on
`urgency_level`	string	Low → Deadline in under 1 hour
`communication_style`	string	Brief imperative, verbose, exploratory, etc.

🤖 `ai_system_response`

Field	Type	Description
`assistant_response`	string	Context-aware AI response
`response_tone`	string	Warm, neutral, energizing, grounding, etc.
`recommended_action`	string	Adaptive productivity recommendation
`ui_adaptation`	string	UI/theme change triggered
`notification_strategy`	string	How notifications are managed
`system_adjustments`	list	OS-level changes applied
`music_recommendation`	string	Adaptive audio environment
`learning_support`	string	Educational strategy offered
`focus_mode`	string	Active focus/productivity mode
`wellness_suggestion`	string	Physical/mental wellbeing nudge

🧬 `long_term_memory`

Field	Type	Description
`historical_behavior_pattern`	string	Pattern observed across sessions
`previous_emotional_state`	string	Prior emotion for trend analysis
`productivity_trend`	string	Multi-session productivity trajectory
`burnout_risk`	string	Negligible → Critical
`learning_progression`	string	Learning trajectory description

📊 Dataset Statistics

Metric	Value
Total Samples	500
Unique Emotions	28
Countries Represented	25
Professions Covered	30
MBTI Personality Types	16 (all)
Age Range	16 – 58
Time Slots	9 (4 AM → 5 AM next day)
Unique App References	34
Total JSON Size	~1.86 MB
Splits	Train / Validation / Test

Emotion Distribution (Top 10)

Emotion	Count
sad	28
calm	25
anxious	25
curious	24
lonely	22
optimistic	21
determined	20
mentally_exhausted	20
restless	20
insecure	20

🚀 Quick Start

Load with Hugging Face `datasets`

from datasets import load_dataset

dataset = load_dataset("wincode/HESA-v1")
print(dataset)
# DatasetDict({
#     train: Dataset({features: [...], num_rows: 400}),
#     validation: Dataset({features: [...], num_rows: 60}),
#     test: Dataset({features: [...], num_rows: 40})
# })

Load raw JSON

import json

with open("HESA_dataset_v1.json", "r") as f:
    data = json.load(f)

samples = data["samples"]
print(f"Total samples: {len(samples)}")

# Access first sample
s = samples[0]
print(f"Emotion: {s['human_behavior']['emotion']}")
print(f"Stress: {s['user_profile']['stress_level']}")
print(f"AI Response: {s['ai_system_response']['assistant_response']}")

Example: Filter by emotion

burned_out = [s for s in samples if s["human_behavior"]["emotion"] == "burned_out"]
print(f"Burned-out samples: {len(burned_out)}")

# Get AI wellness suggestions for burned-out users
for s in burned_out[:3]:
    print(s["ai_system_response"]["wellness_suggestion"])

Example: Stress vs Typing Speed analysis

import pandas as pd

rows = [{
    "stress": s["user_profile"]["stress_level"],
    "typing_wpm": s["system_state"]["typing_speed_wpm"],
    "error_rate": s["system_state"]["error_frequency_rate"],
    "sleep": s["user_profile"]["sleep_hours"]
} for s in samples]

df = pd.DataFrame(rows)
print(df.groupby("stress")[["typing_wpm", "error_rate"]].mean())

🎯 Intended Use Cases

✅ Recommended Uses

Emotion classification — Train models to detect user emotional state from behavioral signals
Adaptive UI/UX systems — Learn context-aware interface adaptation rules
Intelligent tutoring — Model learning state transitions and educational support strategies
Burnout prediction — Detect early burnout signals from multi-session patterns
Cognitive load estimation — Infer workload from typing speed, error rate, and multitasking signals
Affective computing research — Study emotion-behavior-environment relationships
Synthetic data augmentation — Supplement real HCI datasets
AGI training — Human-grounded context for general-purpose assistants

❌ Out-of-Scope Uses

Real identity inference — This is fully synthetic data; no real persons represented
Surveillance systems — Not intended for real-time monitoring applications
Clinical diagnosis — Not a substitute for clinically validated instruments

⚙️ Generation Methodology

HESA-v1 was generated using a Python-based synthetic data engine with:

Psychologically consistent derivation rules — typing speed, error frequency, and frustration signals are mathematically derived from stress level, sleep hours, and emotional state
Realistic variability — Gaussian distributions for focus duration, weighted distributions for sleep hours
Cross-field coherence — Emotion states correlate with UI adaptations, notification strategies, and wellness responses
Diverse seeding pools — 30 professions, 25 countries, 16 MBTI types, 28 emotion categories, 34 real applications
Temporal realism — Timestamps span full year with all time-of-day slots represented

Derived relationships implemented:

stress_level + sleep_hours + mental_focus → typing_speed_wpm
stress_level + sleep_hours + emotion → error_frequency_rate
emotion + emotion_intensity → frustration_signals
emotion + stress + focus → ai_system_response

📁 Repository Structure

wincode/HESA-v1/
├── README.md                    ← This file
├── HESA_dataset_v1.json         ← Full raw dataset (500 samples)
├── data/
│   ├── train.json               ← 400 training samples
│   ├── validation.json          ← 60 validation samples
│   └── test.json                ← 40 test samples
└── notebooks/
    └── explore_HESA.ipynb       ← Exploration notebook (optional)

📜 Citation

If you use HESA-v1 in your research or projects, please cite:

@dataset{hesa_v1_2026,
  title     = {HESA-v1: Human Emotion and System Adaptation Dataset},
  author    = {Shajahan, Ahmad},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/wincode/HESA-v1},
  note      = {Synthetic research-grade dataset for emotion-aware AI systems}
}

📄 License

This dataset is released under the MIT License. Free for academic, research, and commercial use with attribution.

👤 Author

Ahmad Shajahan (wincode)

🤗 HuggingFace: huggingface.co/wincode
🐙 GitHub: github.com/ahmadshajahan
🏢 Zerath Labs / The Art of Engineering — Kerala, India

KTU S4 CSE Student | AI/ML Researcher | PyTorch Content Creator

Made with 🧠 for the future of human-centered AI