README.md

Re-coding Trust — Computational Research Engine

CUHK COMM4150 FYP · 2025–2026
Quantifying Hong Kong's crypto "Trust Paradox": why users trust HashKey Exchange but won't use it.

Key Finding: HashKey users score asset security trust at M=5.27/7 — yet brand relatability at M=2.75/7. The Δ=2.52 gap is statistically significant (t(286) = 24.41, p < .001). This repo measures that gap and informs the campaign strategy →

---

Campaign Deliverables

🍕 Pizza Personas Quiz  ·  🪦 Rest in Blockchain Memorial  ·  📦 hkx CLI

---

Research Architecture: 3-Layer Method

Layer	Method	Scale
§2.1 Computational Social Listening	NLP pipeline — TextBlob (EN) + SnowNLP (ZH)	5,000+ interactions across Twitter/X, LIHKG, Telegram
§2.2 Primary Quantitative Survey	Google Forms, purposive + snowball sampling	N=287 valid responses (18–35, HK-resident)
§2.3 Competitive Benchmark	LBank internal user research (Apr 2025)	N=1,751
§6 Pre-Launch Simulation	DeepSeek-V3 agent-based scenario model	3 scenarios × 2 agent types

---

Key Findings

The Trust Paradox — Quantified (§2.2)

Trust (M=5.27) vs. Relatability (M=2.75): Δ=2.52, t(286)=24.41, p<.001

Pain Point Mapping (§2.1.1)

"Lack of Degen Culture" is the #1 barrier to licensed exchange adoption (33.8%), outranking KYC complexity (27.5%) and high fees (21.6%).

Pizza Day Sentiment Anomaly (§2.1.1)

May 22 is the single highest positive sentiment engagement anomaly in the HK Web3 calendar.

KOL Matrix (§2.1.1)

Mapping influencer reach against cultural alignment (Compliance ↔ Degen) to identify "Translator" voices.

Exchange Usage (§2.2.4)

64.5% primarily use offshore platforms (Binance/OKX/Bybit); only 15.7% use licensed exchanges.

Exchange Selection Factors (§2.2.4)

Regulatory compliance ranked last (M=3.16/5); UX ranked first (M=4.21/5). Compliance is a hygiene factor, not a pull factor.

Main Adoption Barriers (§2.2.4)

Cultural mismatch outranks structural friction.

Bitcoin Pizza Day Awareness (§2.2.4)

66.2% combined awareness; 67.1% among the 18–24 primary cohort.

Preferred Campaign Incentive (§2.2.4)

Token airdrops (45.6%) dominate as the conversion trigger — validating the campaign's on-chain mechanics.

HashKey Word Association (§2.2.4)

"Reliable" (48.1%) and "Safe" (46.7%) co-occur with "Boring" (39.4%) — the Trust Paradox in one chart.

---

Project Structure

.
├── data/
│   ├── raw/
│   │   ├── social_feeds/           # Archived social feeds (500+ records)
│   │   └── survey_raw_responses.csv
│   ├── processed/                  # NLP-analyzed sentiment datasets
│   └── simulation/
│       ├── results.json            # §6 pre-launch simulation output (seed: 4150)
│       └── charts/                 # Fig 6.1–6.4 (generated)
├── docs/                           # Academic methodology documentation
├── outputs/                        # All generated charts (10 files)
└── src/
    ├── collectors/                 # Data ingestion pipelines
    ├── analysis/                   # NLP processor
    ├── visualization/
    │   ├── engine.py               # Social listening charts
    │   └── survey_charts.py        # Survey charts (7 outputs)
    ├── simulation/
    │   ├── campaign_sim.py         # DeepSeek-V3 agent simulation
    │   └── generate_charts.py      # Fig 6.1–6.4 generator
    └── main.py                     # Pipeline orchestrator

---

Tech Stack

Layer	Tools
Language	Python 3.9+
NLP	TextBlob (EN), SnowNLP (ZH), Jieba
Analytics	Pandas, NumPy, SciPy
Visualization	Matplotlib, Seaborn, WordCloud
Simulation	DeepSeek-V3 API (deepseek-chat)

---

Run

pip install -r requirements.txt

# Generate all survey charts (10 outputs)
python src/visualization/survey_charts.py

# Run full social listening pipeline
python src/main.py

# Re-run simulation and regenerate Figs 6.1–6.4
python src/simulation/campaign_sim.py
python src/simulation/generate_charts.py

---

ZHAO Han · CUHK COMM4150 Final Year Project · 2025–2026
Campaign deliverables → github.com/Beltran12138/hkx