This repository corresponds to Siying Qian's Honours Research Project "Context-Aware Document Analysis".
The codes are a runnable and extensional version of codes in "GPT-GNN: Generative Pre-Training of Graph Neural Networks".
Raw data for the pre-processing stage is the data provided in the GPT-GNN repository.
Chosen custom graphs from preprocessing (preprocess_output), pre-trained models (models) and fine-tuned models (finetune_models) mentioned in the "Results Analysis" section of my thesis can be achieved via this link.
Different content analysis versions for titles and abstracts and various keyword extraction versions are listed in the preprocess.py file. Whether abstract embedding is considered corresponds to two GNN model versions in the pretrain.py, finetune_PV.py, finetune_AD.py and GPT_GNN/utils.py files.