Practical overview
When it fits
Use WordGCN when an NLP experiment needs embeddings that incorporate syntactic dependency context, or when refining pretrained SynGCN embeddings with synonym information.
Who it is for
- NLP researchers studying dependency-informed word representations.
- Practitioners comparing syntactic-context embeddings with other word-embedding methods.
- Researchers reproducing the project's embedding-training and evaluation workflows.
Documented uses
- Train word embeddings by predicting words from graph-encoded dependency context.
- Fine-tune pretrained SynGCN embeddings with synonym information.
- Evaluate learned embeddings using word-embedding-benchmarks and described downstream task models.
Requirements & limits
The repository is public and licensed under Apache-2.0.
- The documented environment is TensorFlow 1.x with Python 3.x.
- Training requires a processed corpus with token and dependency-relation data, plus compilation of C++ batch-generation code.
- The README cautions that SynGCN training is slower than the cited word2vec implementation; very large corpora may need multiple GPUs or a C++ implementation.
- The repository documents distinct stages for SynGCN training and SemGCN semantic fine-tuning.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Train SynGCN embeddings from the processed Wikipedia corpus
Prepare the documented corpus and start SynGCN training to save embeddings.
Before you start
- Python 3.x and TensorFlow 1.x
- Project dependencies installed from requirements.txt
- The processed Wikipedia corpus extracted into ./data
- C++ build tools to compile the batch-generation code
Steps
Download and extract the processed Wikipedia corpus into ./data.
Run make to compile the C++ code used for creating batches.
Run syngcn.py with a name, GPU selection, maximum sentence length, maximum dependency length, and embedding dimension.
Expected outcome
The trained embeddings are saved in ./embeddings under the supplied name.
Guide limitations
- The README reports that the TensorFlow implementation is slower than the cited word2vec implementation.
- Training on very large corpora may require multiple GPUs or a C++ implementation.
Fine-tune pretrained embeddings with semantic information
Use SemGCN to incorporate synonym information into pretrained 300-dimensional SynGCN embeddings.
Before you start
- Python 3.x and TensorFlow 1.x
- Project dependencies installed from requirements.txt
- Pretrained 300-dimensional SynGCN embeddings
Steps
Obtain the pretrained SynGCN embeddings described in the README.
Run semgcn.py with the embedding path, semantic setting synonyms, embedding dimension 300, output name, dump option, and GPU selection.
Expected outcome
Fine-tuned embeddings are saved in ./embeddings under the supplied name.
Guide limitations
- The documented example uses pretrained 300-dimensional SynGCN embeddings and the synonym semantic setting.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- malllabiisc/WordGCN · Official homepage title · retained in its source language