📚 Knowledge Graph Construction with Personality Modeling

LLM Sessions : https://gemini.google.com/share/4bf368e3dbbf

Automated pipeline for extracting knowledge graphs from unstructured text with Big Five personality trait inference

---

🎯 Overview

This project implements an intelligent NLP pipeline that automatically constructs Knowledge Graphs (KGs) from unstructured text documents. The system goes beyond basic entity and relationship extraction by inferring and modeling Big Five (OCEAN) personality traits of individuals mentioned in the text.

Key Capabilities:

• 🔍 Entity Recognition & Linking
• 🔗 Relationship Extraction (SPO triples)
• 🧠 Personality Trait Inference (Big Five/OCEAN)
• 📊 Comprehensive Evaluation Metrics
• 🎨 Interactive Graph Visualization

---

✨ Features

🔤 Text Processing

• Supports plain text files (.txt)
• Multi-document batch processing
• UTF-8 and Latin-1 encoding support

🎯 Entity Recognition

• NER Engine: spaCy en_core_web_sm model
• Entity Types: PERSON, ORG, GPE, PRODUCT, LOCATION
• Smart Detection: Handles proper nouns and capitalized names

🔗 Entity Linking & Coreference Resolution

• Unifies name variations (e.g., "Dr. Vance" → "Elias Vance")
• Resolves pronouns to canonical names (He/She → Person)
• Handles titles, prefixes, and possessives
• Advanced substring matching for compound names

📝 Relation Extraction

Uses spaCy's dependency parser to extract:

• Subject-Predicate-Object (SPO) triples
• IS_A relationships (e.g., "Alice is a scientist")
• OWNS relationships (possessive constructs)
• HAS_TRAIT personality connections
• Verb-based and prepositional relationships

🧠 Personality Modeling

Infers Big Five (OCEAN) personality traits:

• Openness to Experience
• Conscientiousness
• Extraversion / Introversion
• Agreeableness
• Neuroticism

Scoring: 1-5 scale based on configurable keyword matching with contextual evidence tracking.

📊 Evaluation Framework

Intrinsic Metrics (No Ground Truth Required)

• Graph Coherence (95.4%): Structure quality and connectivity
• Information Richness (100%): Extraction efficiency per sentence
• Semantic Completeness (95.4%): Personality profile completeness
• Entity Recognition (85%): Detection quality
• Relationship Diversity (86%): Variety of relationship types
• Overall Score: 94.1/100 ⭐⭐⭐⭐⭐

Supervised Metrics (With Ground Truth)

• Entity Precision/Recall/F1-Score
• Relation Precision/Recall/F1-Score (with partial matching)
• Matched vs Expected relations analysis

See EVALUATION_METRICS.md for detailed documentation.

🎨 Visualization

• Color-coded nodes: Person (blue), Trait (green), Concept (gold), etc.
• Labeled edges: Relationship types with scores
• Responsive layout: Spring-layout algorithm for optimal positioning

---

🛠️ Technologies

Technology	Purpose	Version
Python	Core language	3.7+
spaCy	NLP (NER, POS, parsing)	3.0+
NetworkX	Graph construction	Latest
Matplotlib	Static visualization	Latest
Pandas	Evaluation reports	Latest

---

🚀 Quick Start

📹 Quick Demo

# Clone the repository
git clone https://github.com/DerickAmadeus/Knowledge-Graph-Challenge.git
cd Knowledge-Graph-Challenge

# Install dependencies
pip install spacy networkx pandas matplotlib

# Download spaCy model
python -m spacy download en_core_web_sm

# Run the pipeline
python main.py
and the result is automatically popped up

1️⃣ Prerequisites

# Ensure Python 3.7+ is installed
python --version

# Install required packages
pip install spacy networkx pandas matplotlib

2️⃣ Download spaCy Model

python -m spacy download en_core_web_sm

3️⃣ Prepare Input Files

Create text files with your documents:

project/
├── main.py
├── text
     ├── 1.txt          # Your first document
     ├── 2.txt          # Your second document
     └── 3.txt          # Your third document

Example content (1.txt):

Alice is a kind and brilliant scientist who works with Bob in London. 
She often leads meetings and enjoys collaborating with her colleagues.

4️⃣ Run the Pipeline

python main.py

5️⃣ View Results

Console Output:

[GRAPH STRUCTURE]
  * Total Nodes: 10
  * Total Edges: 13
  * Persons: 2
  * Trait Coverage: 100.0%

[INTRINSIC QUALITY METRICS]
  * Graph Coherence: 95.4%
  * Information Richness: 100.0%
  * Intrinsic Score: 94.1/100

⚙️ Configuration

Personality Keywords

Edit DEFAULT_TRAIT_CLUES in main.py:

DEFAULT_TRAIT_CLUES = {
    "Conscientiousness": {
        "keywords": ["focused", "methodical", "organized", "diligent"],
        "score_mod": 1.0
    },
    "Extraversion": {
        "keywords": ["leads meetings", "networking", "talkative"],
        "score_mod": 1.0
    },
    # Add your custom keywords...
}

Title Prefixes

Customize name cleaning patterns:

TITLE_PREFIXES = r'(Dr\.|Prof\.|Professor|Mr\.|Ms\.|Mrs\.|President|CEO)'

Ground Truth (Optional)

Define expected entities and relations for supervised evaluation:

GROUND_TRUTH = {
    "1.txt": {
        "persons": ["Alice", "Bob"],
        "relations": [
            ("Alice", "IS_A", "scientist"),
            ("Alice", "HAS_TRAIT", "Openness"),
        ]
    }
}

---

📊 Example Output

Knowledge Graph Visualization

Evaluation Report

File: 1.txt
─────────────────────────────────────────
Graph Coherence:      95.4% ✓
Information Richness: 100.0% ✓
Semantic Completeness: 95.4% ✓
Entity Recognition:   85.0% ✓
Relationship Diversity: 86.0% ✓
─────────────────────────────────────────
INTRINSIC SCORE: 94.1/100 ⭐⭐⭐⭐⭐

---

📁 Project Structure

Intellumina/
├── main.py                    # Main pipeline script
├── 1.txt, 2.txt, 3.txt       # Input documents
├── knowledge_graph_*.html     # Generated visualizations
├── evaluation_report.csv      # Metrics summary
├── EVALUATION_METRICS.md      # Detailed metrics documentation
├── README.md                  # This file
└── lib/                       # JavaScript libraries
    ├── vis-9.1.2/
    └── tom-select/

---

🔬 Evaluation Metrics

Without Ground Truth (Intrinsic)

Perfect for production use when you don't have annotated data:

Metric	Weight	Description
Graph Coherence	20%	Structural quality, connectivity
Information Richness	25%	Extraction efficiency
Semantic Completeness	30%	Personality profile completeness
Entity Recognition	15%	Detection quality
Relationship Diversity	10%	Variety of relations

With Ground Truth (Supervised)

For research and benchmarking:

• Entity Precision/Recall/F1
• Relation Precision/Recall/F1 (partial matching supported)
• Comprehensive accuracy analysis

📖 Full documentation: EVALUATION_METRICS.md

---

🎯 Use Cases

• 📝 Research Papers: Extract author relationships and expertise
• 💼 Business Documents: Identify key stakeholders and roles
• 📰 News Articles: Map people, organizations, and events
• 🎓 Academic Analysis: Study personality traits in literature
• 🔍 HR & Recruitment: Analyze candidate profiles
• 🤝 Social Network Analysis: Understand relationship dynamics

---

⚠️ Limitations & Future Work

Current Limitations

Area	Limitation	Impact
Relation Extraction	Rule-based dependency patterns	May miss complex/implicit relations
Coreference	Heuristic name mapping	Occasional fragmentation
Personality Model	Keyword matching	Limited context sensitivity
Negation	Not handled	"not organized" → still scores high
Sarcasm	Not detected	Misinterprets ironic statements

---

📧 Contact & Support

• Author: Derick Amadeus Budiono
• Repository: Knowledge-Graph-Challenge
• Issues: Report bugs
• Discussions: Q&A and Ideas

---

📊 Project Statistics

Language: Python
Lines of Code: ~800
Functions: 15+
Evaluation Metrics: 10+
Supported Entities: 7 types
Personality Traits: 5 (Big Five)

---

🌟 Star History

If you find this project useful, please consider giving it a star ⭐!

---

⬆ Back to Top