| name | anti-corruption-investigation |
| description | Advanced anti-corruption investigation system for analyzing chat logs and communications to detect suspicious patterns, corruption indicators, and relationship networks. Supports Chinese and English chat logs in JSON/TXT formats, handles million-scale datasets, and provides human-friendly relationship analysis with evidence-backed conclusions. |
Anti-Corruption Investigation v7.0
What's New in v7.0
Multi-Language Pattern Detection
- Chinese Patterns: Financial corruption, power abuse, secret meetings, collusion
- English Patterns: Evidence destruction, insider trading, pressure manipulation, enterprise fraud
- Universal Enterprise Fraud Detection: SPE, accounting manipulation, financial metrics manipulation
- Semantic Pattern Matching: Detects implicit expressions and euphemisms
Social Network Analysis
- Person Profile Analysis: Comprehensive profiling including role detection, activity patterns, and risk assessment
- Intermediary Detection: Automatically identifies bridge persons connecting corruption networks
- Community Detection: Discovers corruption groups based on communication patterns
- Influence Analysis: Ranks individuals by network influence and centrality
- Connection Path Analysis: Finds shortest paths between high-risk individuals
Validation Framework
- Pattern Validation: Validates detection accuracy against known corruption patterns
- Report Validation: Ensures report completeness and quality
- False Positive Control: Estimates and controls false positive rates
- Continuous Improvement: Generates recommendations for pattern enhancement
When to Use This Skill
Use when analyzing chat logs, messages, or communications for:
- Corruption detection: Financial corruption, power abuse, secret meetings, collusion
- Relationship analysis: Identifying key players, corruption networks, intermediaries
- Social network analysis: Understanding person profiles, influence, and group structures
- Large-scale analysis: Processing 100K+ messages efficiently
- Evidence gathering: Extracting specific evidence for relationships
- Risk assessment: Evaluating corruption risk levels
Quick Start
Basic Analysis
from anti_corruption import ChatAnalyzer
messages = [...]
analyzer = ChatAnalyzer(messages)
results = analyzer.analyze()
print(f"Risk Level: {results['risk_level']}")
print(f"Suspicious Messages: {len(results['suspicious_messages'])}")
Multi-Language Pattern Detection
from multi_lang_patterns import analyze_text, analyze_email
result = analyze_text("We need to delete these documents before audit")
print(f"Risk Score: {result['risk_score']}")
print(f"Categories: {result['categories']}")
email_data = {
'sender': 'john@company.com',
'receiver': 'jane@company.com',
'subject': 'Q4 Results',
'content': 'We need to hit the target number.',
'title': 'CFO'
}
result = analyze_email(email_data)
print(f"Risk Level: {result['risk_level']}")
Relationship Analysis
from anti_corruption import RelationshipAnalyzer
analyzer = RelationshipAnalyzer(messages)
relationships = analyzer.analyze()
for rel in relationships['top_relationships'][:10]:
print(f"{rel['person_a']} ↔ {rel['person_b']}")
print(f" Type: {rel['relationship_type']}")
print(f" Evidence: {len(rel['evidence'])} items")
print(f" Risk: {rel['risk_level']}")
Social Network Analysis
from anti_corruption import SocialNetworkAnalyzer
analyzer = SocialNetworkAnalyzer(messages)
results = analyzer.analyze()
for name, profile in results['person_profiles'].items():
print(f"{name}: {profile['primary_role']} - {profile['risk_level']}")
for inter in results['intermediaries'][:5]:
print(f"Intermediary: {inter['name']} (Score: {inter['brokerage_score']})")
for comm in results['communities']:
print(f"Community: {', '.join(comm['members'][:5])}")
Validation
from case_validator import validate_analysis, generate_validation_report
validation = validate_analysis(analysis_results)
print(f"Detection Accuracy: {validation.detection_accuracy:.1%}")
print(f"False Positive Rate: {validation.false_positive_rate:.1%}")
report = generate_validation_report(validation, 'validation_report.txt')
Core Scripts
anti_corruption.py
Unified analysis tool with all features.
Usage:
python anti_corruption.py analyze input.jsonl report.json
python anti_corruption.py relationships input.jsonl relationships.json --text-report report.txt
python anti_corruption.py social-network input.jsonl social_network.json --text-report social_report.txt
python anti_corruption.py full input.jsonl output_dir/
Commands:
analyze: Basic corruption pattern detectionrelationships: Relationship network analysissocial-network: Social network and person profile analysisfull: Run all analyses
multi_lang_patterns.py
Multi-language pattern detection module.
Features:
- Automatic language detection (Chinese, English, Mixed)
- Direct pattern matching for corruption indicators
- Semantic pattern matching for implicit expressions
- Enterprise fraud specific patterns (SPE, accounting manipulation)
- Risk scoring based on pattern matches
Usage:
from multi_lang_patterns import MultiLangPatternMatcher, EnterpriseFraudDetector
matcher = MultiLangPatternMatcher()
matches = matcher.match_patterns(text)
summary = matcher.get_summary(matches)
detector = EnterpriseFraudDetector()
result = detector.analyze_email(email_data)
case_validator.py
Validation framework for analysis results.
Features:
- Pattern detection validation
- False positive rate estimation
- Report completeness checking
- Improvement recommendations
Usage:
from case_validator import PatternValidator, ReportValidator
validation = PatternValidator.validate_analysis(results)
report_check = ReportValidator.validate_report(report)
Data Format
Input Format (JSONL)
{"timestamp": "2024-01-15T14:30:00", "sender": "张三", "receiver": "李四", "content": "那笔钱准备好了吗？"}
{"timestamp": "2024-01-15T14:31:00", "sender": "李四", "receiver": "张三", "content": "已经准备好了"}
Input Format (TXT)
[2024-01-15 14:30:00] 张三 -> 李四: 那笔钱准备好了吗？
[2024-01-15 14:31:00] 李四 -> 张三: 已经准备好了
Output Format
Social Network Analysis Output
{
"person_profiles": {
"张三": {
"name": "张三",
"message_count": 150,
"contact_count": 8,
"contacts": ["李四", "王五", ...],
"primary_role": "official",
"detected_roles": ["official", "business"],
"suspicious_message_count": 25,
"corruption_patterns": {
"financial_corruption": 15,
"power_abuse": 10
},
"risk_score"
Pattern Categories
Financial Corruption
- Chinese: 转账, 汇款, 回扣, 贿赂, 好处费
- English: kickback, bribe, hidden payment, secret fee
Evidence Destruction
- Chinese: 删除, 销毁, 清理, 不留痕迹
- English: delete, destroy, shred, clean up, off the record
Insider Trading
- English: stock option, insider information, before announcement
Enterprise Fraud
- SPE, off-balance sheet, mark-to-market
- Aggressive accounting, earnings management
- EBITDA manipulation, pro forma adjustments
Pressure Manipulation
- English: pressure, hit the target, make it happen
- Adjust numbers, bridge the gap, find a way
Version History
- v7.0: Added multi-language pattern detection, enterprise fraud patterns, validation framework
- v6.0: Added social network analysis, person profiling, intermediary detection
- v5.0: Refactored for clarity, human-friendly output, improved performance
- v4.0: Added relationship network analysis
- v3.0: Large-scale processing support
- v2.0: Semantic pattern matching
- v1.0: Initial release with keyword-based detection
