search

DavideBuffelli / SAME

Code for the papers: "Graph Representation Learning for Multi-Task Settings: a Meta-Learning Approach", "A Meta-Learning Approach for Graph Representation Learning in Multi-Task Settings"
Other
16 stars 5 forks source link

can be implemented to different dataset? #3

Closed aozorahime closed 1 year ago

aozorahime commented 1 year ago

Hi, I read your paper and it is totally awesome about MTL for graph representation learning. I wonder whether I can use this for text classification since your dataset is mostly graph-structured whereas in my case I shall convert the text to the graph by using GCN (for exmaple). Need your explanation. Thanks.

DavideBuffelli commented 1 year ago

Hi, I think there are two ways in which you can do this: 1 - Obtain a graph representation of your data (e.g., creating parse graphs for each phrase), and then put it into a PyTorch Geometric dataset so that you can use it with this code. As long as you implement this method (https://github.com/DavideBuffelli/SAME/blob/cfbdda3a4b63bb4a5b92ec6cd21a37fcede47bad/data_utils.py#L41) such that it returns a PyTorch Geometric dataset, the code should work 2 - Keep your data as text, and use a different architecture instead of a GCN