GraphTopic

Graph-based topic modeling built on semantic graphs.

GraphTopic is a Python library for discovering document topics using graph-based community detection instead of traditional clustering algorithms.

It is built on top of Graphion, a modular graph processing framework, allowing every stage of the topic modeling pipeline to be customized.

Features

• Graph-based topic discovery
• Pluggable embedding models
• Pluggable graph construction
• Pluggable community detection
• Pluggable topic representation
• Simple sklearn-like API
• Built on Graphion

Installation

Install the core library:

pip install graphtopic

To use the built-in SentenceTransformer embedding model:

pip install "graphtopic[embedding]"

Quick Start

from graphtopic import GraphTopic
from graphtopic.embedding_models import (
    SentenceTransformerEmbedding,
)

documents = [
    "Apple released a new iPhone.",
    "Samsung announced a new Galaxy phone.",
    "Barcelona won the football match.",
    "Real Madrid signed a new player.",
]

model = GraphTopic(
    embedding_model=SentenceTransformerEmbedding(
        "sentence-transformers/all-MiniLM-L6-v2"
    )
)

topics = model.fit_transform(documents)

for topic in model.get_topic_info():
    print(topic)

Pipeline

GraphTopic follows a modular processing pipeline:

Documents
↓
Embeddings
↓
FeatureSet
↓
Relation Graph
↓
Community Detection
↓
Topic Representation
↓
Topic Information

Each stage can be replaced with custom implementations.

Custom Components

GraphTopic is designed around interchangeable components.

You can replace:

• Embedding Model
• Reducer
• Relation Builder
• Graph Builder
• Graph Refiner
• Community Detector
• Partition Refiner
• Representation Model

This makes it easy to experiment with different graph construction and topic discovery strategies.

Built on Graphion

GraphTopic uses Graphion as its graph processing backend.

Graphion provides reusable components for:

• similarity computation
• graph construction
• graph refinement
• community detection
• graph processing pipelines

GraphTopic focuses only on topic modeling while delegating graph operations to Graphion.

Requirements

• Python 3.11+
• NumPy
• scikit-learn
• Graphion

Project Status

GraphTopic is currently in Pre-Alpha (0.0.x).

The public API is expected to evolve as new functionality is added.

License

MIT License