Engine Active v2.4
GitHub

๐Ÿ” Account Forensic Auditor

Multi-dimensional analytical telemetry analyzer evaluating follower asymmetry, circadian entropy, cron-like inter-arrival timing, and linguistic repetition.

Target Profiles (Pre-Calibrated Case Studies) Click any profile to audit
โš™๏ธ Account Parameters MANUAL / API
97.4%
Bot Probability
๐Ÿšจ High Risk Bot Twitter / X Spam & Phishing

Automated Sybil / Spam Bot Detected

This account exhibits severe bot characteristics: extreme follower/following asymmetry (14:4890), excessive posting frequency (142 posts/day), absence of circadian sleep rhythm, and clockwork temporal intervals.

๐Ÿ“Š Forensic Feature Vectors NORMALIZED [0-1]
Follower Reputation Score 0.003 (Critical)
Post Frequency / Velocity 142.6 posts/day
Temporal Uniformity (Cron Indicator) 0.04 (Clockwork)
Circadian Sleep Entropy (24/7 Activity) 0.98 (Non-Stop)
Lexical Repetition (Type-Token Ratio) 0.18 (Spam Template)
Username Shannon Entropy 3.12 bits (Alphanumeric)
๐Ÿง  Explainable AI (SHAP Attribution) LOCAL IMPACT
โฐ 24-Hour Circadian Activity Distribution UTC POSTING PATTERN
00:00 (Midnight) 06:00 (Dawn) 12:00 (Noon) 18:00 (Dusk) 23:59
Sample Posts Telemetry Stream

๐Ÿ“Š Batch Benchmark Studio

Comprehensive machine learning model validation across 120+ synthetic & empirical social media account records based on Cresci-2017 & Botometer protocols.

Dataset Cohort ๐Ÿ“
120
Balanced Bot / Human Baselines
Model Accuracy ๐ŸŽฏ
98.3%
Overall Correct Classifications
Precision Rate ๐Ÿ”ฌ
98.3%
Low False Positive Rate
Recall / Detection ๐Ÿ›ก๏ธ
100.0%
Bot Identification Recall
F1-Score โšก
0.991
Harmonic Mean Metric
๐Ÿงฎ Confusion Matrix BINARY (BOT vs HUMAN)
60 True Positive (TP)
Bots correctly flagged
1 False Positive (FP)
Humans flagged as bot
0 False Negative (FN)
Bots classified as human
59 True Negative (TN)
Humans correctly cleared
* Optimal cutoff configured at 50% Risk Threshold. Minimizes false positives to prevent suspending legitimate users.
๐Ÿ“ˆ Risk Score Distribution Histogram BIMODAL SEPARATION
Clear bimodal separation illustrates high model discriminative power between legitimate organic users (left peak < 20%) and automated bots (right peak > 80%).
0% (Human) 25% 50% (Cutoff) 75% 100% (Bot)
๐Ÿ“‹ Benchmark Accounts Ledger
Handle Platform Followers / Following Age Posts/Day Ground Truth Predicted Risk Verdict Action

๐Ÿ•ธ๏ธ Sybil Botnet & Astroturfing Visualizer

Interactive 2D physics-based network graph mapping follower-rings, coordinated astroturfing clusters, and malicious bridge nodes attempting infiltration.

Cluster Topology
Cluster Alpha: Crypto Spam Ring
Cluster Beta: Sybil Astroturf Ring
Cluster Gamma: Brand Phishers
Cluster Delta: Stanford / Tech Organic
Infiltration Sleeper Bridges
Node Telemetry

@crypto_pump_bot99

High Risk Bot (97%)
Node acts as a high-degree coordinated central hub for 6 worker airdrop amplification nodes.

โš™๏ธ Model Laboratory & Heuristic Tuner

Calibrate feature weights and risk thresholds. Observe live changes to false-positive rates and test custom synthetic profiles in real time.

๐ŸŽ›๏ธ Heuristic Risk Weights
Posting Velocity Penalty Weight 22.0
Clockwork Inter-Arrival Timing Weight 24.0
Circadian Rhythm Absence Weight 16.0
Reputation Asymmetry Weight 18.0
Decision Threshold Cutoff (%) 50%
Real-time Impact on Benchmark:
Updated Accuracy: 98.3% False Positives: 1
๐Ÿงช Custom Synthetic Profile Simulator LIVE PREVIEW
Followers / Following Ratio 10 / 2500 (0.004)
Posting Volume (Posts / Day) 120 posts/day
Temporal Inter-Arrival Variance 0.05 (Cron Automation)
Circadian 24/7 Sleep Entropy 0.92 (Continuous Activity)
Simulated Result
94.8%
๐Ÿšจ High Risk Bot Classification

๐Ÿ“‘ Analytical Detection Methodology

Scientific foundations, mathematical models, feature engineering taxonomy, and countermeasures against adversarial social media manipulation.

1. Executive Overview & Problem Definition

Automated social media accounts ("bots"), sybil networks, and astroturfing campaigns distort public consensus, facilitate phishing scams, and degrade user trust across modern digital platforms. This project introduces a multi-tier analytical detection architecture combining static metadata indicators, non-stationary temporal Poisson dynamics, linguistic entropy, and graph topological metrics.

2. Mathematical Formulation of Detection Signals

A. Follower-Following Reputation Index ($R$)

Legitimate human profiles cultivate bi-directional social graphs. Mass-following spam bots exhibit extreme asymmetry:

R = Followers / (Followers + Following + ฮต) where R โˆˆ [0, 1]. R < 0.05 indicates acute asymmetric bot behavior.

B. Shannon Entropy of Alphanumeric Handles ($H(X)$)

Synthetic usernames generated via PRNG or batch scripts exhibit distinct character entropy relative to natural human phonetic phonemes:

H(X) = - ฮฃ [ P(x_i) ยท logโ‚‚(P(x_i)) ] where P(x_i) is the probability frequency of character x_i in handle string X.

C. Inter-Arrival Time Variance (Poisson Process Analysis)

Organic human interactions are modeled by bursty Pareto or non-homogeneous Poisson processes with high inter-arrival interval variance $\sigma^2(\Delta t)$. Conversely, automated cron scripts emit events at near-zero variance intervals ($\Delta t \approx \text{const}$):

ฯƒยฒ(ฮ”t) = (1 / (N - 1)) ยท ฮฃ (ฮ”t_i - ฮผ_ฮ”t)ยฒ Clockwork threshold: ฯƒยฒ(ฮ”t) < 0.15 triggers automated scheduling penalty.

D. Circadian Sleep Rhythm Entropy

Human biological circadian rhythms dictate 6โ€“8 hour periods of near-zero posting activity. The circadian entropy metric measures the uniform dispersion of events across 24 hourly buckets:

CircadianEntropy = - (1 / logโ‚‚(24)) ยท ฮฃ [ p_h ยท logโ‚‚(p_h) ] for h โˆˆ [0, 23] Continuous 24/7 posting yields CircadianEntropy โ†’ 1.0 (Automated Farm).

3. Explainable AI (SHAP-Inspired Local Attribution)

To maintain transparency for trust and safety operators, the decision engine computes linear additive feature attributions:

g(z') = ฯ†โ‚€ + ฮฃ ฯ†_i ยท z'_i where ฯ†_i represents the positive or negative risk delta contributed by signal i.

4. Defensive Countermeasures & Sybil Mitigation

Based on the analytical scores produced by CyberSentry AI, platforms deploy graduated countermeasures:

  • Score < 30%: Verified Authentic. Unrestricted platform access.
  • Score 30% - 65%: Suspicious. Enforce dynamic proof-of-work, rate limits, and SMS/TOTP step-up verification.
  • Score > 65%: High Risk. Shadowban feed distribution, trigger CAPTCHA challenges, and submit to human forensic review.
  • Score > 90%: Malicious Sybil Node. Immediate cryptographic API token revocation and IP cluster quarantine.