Knowledge Graph¶

GraphMem automatically builds a knowledge graph from your text.

What Gets Extracted¶

Entities¶

Type	Examples
Person	Elon Musk, Dr. Smith, CEO
Organization	Tesla, Google, Anthropic
Product	GPT-4, Model S, iPhone
Location	Austin, California, HQ
Date/Time	2024, Q3, January
Number	$35B, 26,000, 94%
Concept	AI, sustainability, revenue

Relationships¶

Relationship	Example
CEO_OF	Elon Musk → Tesla
FOUNDED	Steve Jobs → Apple
WORKS_AT	Alice → Google
LOCATED_IN	Tesla HQ → Austin
ACQUIRED	Apple → NeXT
CREATED	Anthropic → Claude

Attributes¶

Attribute	Example
Aliases	"Elon", "Musk", "CEO"
Description	"Electric vehicle company"
Properties	revenue, employees, founded

Example Extraction¶

Input Text¶

memory.ingest("""
    In Q3 2024, Nvidia (NVDA) reported $35.1B revenue, up 94% YoY.
    CEO Jensen Huang announced Blackwell B200 shipping in early 2025.
    The company has 26,000 employees worldwide.
""")

Extracted Graph¶

                    ┌─────────────┐
                    │   Nvidia    │
                    │ (Company)   │
                    │ aliases:    │
                    │ NVDA        │
                    └──────┬──────┘
                           │
        ┌──────────────────┼──────────────────┐
        │                  │                  │
        ▼                  ▼                  ▼
┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│Jensen Huang │    │   $35.1B    │    │Blackwell    │
│  (Person)   │    │  (Revenue)  │    │   B200      │
│             │    │  Q3 2024    │    │ (Product)   │
│ CEO_OF ─────┼────┼─────────────│    │ ships: 2025 │
└─────────────┘    │  +94% YoY   │    └─────────────┘
                   └─────────────┘
                          │
                          ▼
                   ┌─────────────┐
                   │   26,000    │
                   │ (Employees) │
                   └─────────────┘

Alias Resolution¶

GraphMem automatically handles entity aliases:

memory.ingest("""
    Dr. Alexander Chen, also known as "The Quantum Pioneer",
    founded Quantum AI Labs. Alex Chen received his PhD from MIT.
""")

All these refer to the same entity: - Dr. Alexander Chen - The Quantum Pioneer - Alex Chen

# All queries work:
memory.query("What did Dr. Chen do?")
memory.query("Who is Alexander Chen?")
memory.query("Tell me about The Quantum Pioneer")
memory.query("What did Alex Chen study?")

Entity Resolution¶

When ingesting multiple documents, GraphMem resolves duplicates:

memory.ingest("Elon Musk is CEO of Tesla")
memory.ingest("Tesla's CEO Musk announced new plans")
memory.ingest("Mr. Musk spoke at the conference")

Result: One "Elon Musk" entity with all relationships, not three duplicates.

Resolution Strategy¶

Exact match: Same name
Alias match: Known alias
Embedding similarity: Similar meaning
LLM verification: Final check for edge cases

Graph Traversal¶

GraphMem can traverse relationships to answer complex questions:

memory.ingest("Apple was founded by Steve Jobs.")
memory.ingest("Steve Jobs also founded NeXT.")
memory.ingest("NeXT was acquired by Apple in 1997.")
memory.ingest("Tim Cook worked at Compaq before Apple.")

response = memory.query("What's the connection between NeXT and Tim Cook?")

Traversal path:

NeXT → Steve Jobs → Apple → Tim Cook

Answer: "NeXT was founded by Steve Jobs and acquired by Apple in 1997. Tim Cook currently works at Apple as CEO."

Community Detection¶

Related entities are grouped into communities:

┌─────────────────────────────────────────┐
│           TECH LEADERSHIP               │
│                                          │
│  Elon Musk ─── Tesla ─── SpaceX         │
│      │                                   │
│  Jensen Huang ─── Nvidia                │
│      │                                   │
│  Tim Cook ─── Apple                     │
└─────────────────────────────────────────┘

┌─────────────────────────────────────────┐
│           AI COMPANIES                   │
│                                          │
│  OpenAI ─── GPT-4 ─── Sam Altman        │
│      │                                   │
│  Anthropic ─── Claude ─── Dario Amodei  │
└─────────────────────────────────────────┘

Communities are used for: - High-level summaries - Cross-cluster queries - Topic identification

PageRank Importance¶

Hub entities (connected to many things) score higher:

# PageRank scores
Elon Musk:      PR = 1.000  ████████████████████  # Hub
Tesla:          PR = 0.774  ███████████████
SpaceX:         PR = 0.774  ███████████████
Austin:         PR = 0.520  ██████████            # Peripheral

Higher PageRank = More likely to be retrieved and weighted in answers.