Linkprediction- is it any graph model? & procedure help!

Hi there !

i’m doing mini-project which is building recommendation engine with graph neural network.

stuck in the problem

  1. my raw data property is ‘user - item’ bipartite graph (interaction)

which is proper graph data type ? (homo vs. hete)

  1. i want to resolve my task using link prediction scenario

i understand ‘negative sampling’ is redefine the label at the dataset.

If above thing is right , then i just follow the user guide at dgl official document ?


I think this is exactly the case in our link prediction cases. You can also take a look on our GCMC model. Did you meet any problem?

Thanks to response @VoVAllen

I’m Doing Graph schema modeling ( adjacency matrix building , Bipartite-graph building etc…)

Is there any tutorial for my problem which is dataset for proper setting at DGL ?!

my dataset format is now here

user | product
0 | p_121

set(user node) ; 11268
set(product.node) ; 1664
interaction ; 323897

and user , product having respectively node feature.

It’s similar to Movielens dataset I think, which is a bipartite graph between user and movie with rating as edges

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.