Skip to main content

A library for building text graphs for the application of Graph Neural Networks (GNN), in the context of text classification in natural language processing

Project description

Text for GCN

GCN applied in a text classification context.

License: MIT PyPI Python 3.8 Documentation Status Open In Colab

Table of Contents

  1. Abstract
  2. Installation
  3. Functionalities
  4. Examples
  5. Contributing
  6. License
  7. References


Text4GCN is an open-source python framework that simplifies the generation of text-based graph data to be applied as input to graph neural network architectures. Text4GCN's core is the ability to build memory-optimized text graphs, using different text representations to create their relationships and define the weights used for edges.

This project aims to exam the text classification problem with novel approaches Graph Convolutional Networks and Graph Attention Networks using Deep Learning algorithms and Natural Language Processing Techniques.

The main contribution of this work is to provide a flexible framework, capable of performing syntactic and semantic filters that make text graphs smaller and more representative. This framework offers an alternative and powerful tool for the study of Convolutional Graph Networks applied in the text classification task.

Text graph


Text4GCN is available at PyPI:

pip install text4gcn

Also, Text4GCN can be cloned directly from GitHub ( and run as a Python script.


  • Datasets: Module responsible for downloading model datasets, used in benchmark tasks for text classification.
  • Preprocess: It performs dataset processing, applies natural language processing to process the information and generates the files necessary for the construction of text graphs.
  • Build Adjacency: Creates the adjacency matrix based on a specific representation.
  • Models: Provides a two-tier GCN built with PyTorch for document classification task.


Get data

from text4gcn.datasets import data

# List of all available datasets

# Download sample data for a specific folder

Available Datasets:

  • R8 (Reuters News Dataset with 8 labels)
  • R52 (Reuters News Dataset with 52 labels)
  • 20ng (Newsgroup Dataset)
  • coming soon Ohsumed (Cardiovascular Diseases Abstracts Dataset)
  • coming soon MR (Movie Reviews Dataset)
  • coming soon Cora (Citation Dataset)
  • coming soon Citeseer (Citation Dataset)
  • coming soon Pubmed (Citation Dataset)

Datasets Description:

Dataset Docs Training Test Words Nodes Classes Average Length
20NG 18,846 11,314 7,532 42,757 61,603 20 221.26
R8 7,674 5,485 2,189 7,688 15,362 8 65.72
R52 9,100 6,532 2,568 8,892 17,992 52 69.82
MR 10,662 7,108 3,554 18,764 29,426 2 20.39
Ohsumed 7,400 3,357 4,043 14,157 21,557 23 135.82

Text Pipeline

from text4gcn.preprocess import TextPipeline

# Create a text pipeline for processing a dataset
pipe = TextPipeline(

# Run the created pipeline

Frequency Adjacency

from text4gcn.builder import FrequencyAdjacency

# Create the adjacency matrix based on a specific builder
freq = FrequencyAdjacency(

# Run the created pipeline

Available Builders:

  • Liwc Linguistic Inquiry and Word Count to extract a dependency relationship
  • Frequency
  • Embedding Based on Word2vec, applied due to its ability to capture semantic information for word representations
  • CosineSimilarity
  • DependencyParsing Based on the Syntactic Dependency Tree extracted with Stanford CoreNLP
  • coming soon ConstituencyParsing


from text4gcn.models import Builder as bd
from text4gcn.models import Layer as layer
from text4gcn.models import GNN

gnn = GNN(
    dataset="R8",           # Dataset to train
    path="my_folder",       # Dataset path
    log_dir="examples/log", # Log path
    layer=layer.GCN,        # Layer Type
    epoches=200,            # Number of traing epoches
    dropout=0.5,            # Dropout rate
    val_ratio=0.1,          # Train data used to validation
    early_stopping=10,      # Stop early technique
    lr=00.2,                # Initial learing rate
    nhid=200,               # Dimensions of hidden layers
    builder=bd.Embedding    # Type of Filtered Text Graph


Contributions are greatly appreciated. If you want to help us improve this software, please fork the repo and create a new pull request. Don't forget to give the project a star! Thanks!

  1. Fork the Project
  2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
  3. Commit your Changes (git commit -m 'Add some AmazingFeature')
  4. Push to the Branch (git push origin feature/AmazingFeature)
  5. Open a Pull Request

Alternatively, you can make suggestions or report bugs by opening a new issue with the appropriate tag ("feature" or "bug") and following our Contributing template.


Distributed under the MIT License. See LICENSE.txt for more information.


  • [Kipf and Welling, 2017] Semi-supervised Classification with Graph Convolutional Networks
  • [Liang Yao, Chengsheng Mao, Yuan Luo, 2018] Graph Convolutional Networks for Text Classification

(back to top)

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

text4gcn-1.0.0.tar.gz (36.2 kB view hashes)

Uploaded source

Built Distribution

text4gcn-1.0.0-py3-none-any.whl (43.2 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Fastly Fastly CDN Google Google Object Storage and Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page