In the paper they apply Graph Neural Networks to text classification.
The plan for the project is to use the ideas from the paper to create an academic paper search.
Planned steps:
Construct a graph with paper, and word nodes (maybe author / publisher). Papers are connected to the words in the paper. Edge weights are tf-idf based. Word-word connections are based on co-occurrence statistics.
Learn embeddings for all entities using a GNN. Maybe use TransE or a similar head.
Enable similarity search with vector db. (e.g. search for similar papers, search papers by keywords, search authors by keywords, get keywords for papers).
It would be an experimental application of the paper to search.
Matthias Fast, Ahmet Korkmaz, Henrik Rathai, Amos Dinh
In the paper they apply Graph Neural Networks to text classification. The plan for the project is to use the ideas from the paper to create an academic paper search.
Planned steps:
It would be an experimental application of the paper to search.
Matthias Fast, Ahmet Korkmaz, Henrik Rathai, Amos Dinh
Repository: https://github.com/AmosDinh/GNNpapersearch