Word embeddings for ogbl-collab

[ChatterjeeAyan]

In ogbl-collab, all nodes come with 128-dimensional features, obtained by averaging the word embeddings of papers that are published by the authors. Got a few questions regarding this:

1. What kind of word embedding is used here? Word2vec?
2. Do you use a pre-trained Word2ve for the purpose? Is that model open-source?
3. Is the corpus of the papers available on which embeddings were created?

[weihua916]

Hi!

1. It is word2vec.
2. The model is trained on academic paper. The model is not open-sourced. It was trained a long time ago, and I do not think we can open-source the model.
3. No, it is not available.