Collaborative Filtering is such an underrated computation method, imho Many real life problems can be re-interpteted as a recommendation problem. It's just a matter of perspective, but the results can be huge!
that s an amazing job! clearly honest deep effort i ncompliling such huge info in a compact and unified language! Congrats on this level of sophistication! Predict an even brighter future for you!
I like your videos so much, so plz keep going on. My question is how can I build a simple model of a recommendation system using graph convolutional network?
how did you find these papers? I'm interested in developing a recommendation system for creating outfits for a person i.e. you add a shirt to your shopping cart, the recommender will show pants, shoes, and belts that match the shirt
I have a video on node classification :) it's pretty much exactly that, only that the head of the network is different. Instead of one output per node, you have one output per possible edge :)
I must say your videos are great please keep up the good work. My question is how can i implement a simple recommender system with pytorch geometric. I have a custom dataset with two nodes: movies and user. Furthermore there is an edge that describes only the relation of both nodes. I want to create a bipartite graph like the one from kipf & welling and do a similarity analysis. I know how to get the data into the right format. But since I am new to the subject, I am having a hard time implementing a simple gnn model. If anyone here could help me I would be very grateful.
Hi! There is a section on bipartite graphs in Pytorch geometric: pytorch-geometric.readthedocs.io/en/latest/notes/batching.html?highlight=Bipartite#bipartite-graphs Also I have a couple of videos on how to create custom datasets. Let me know if you need further help!
@@DeepFindr Hello, thank you for the quick feedback. I have now watched many of your videos again. I know how to create a bipartite garph with Heterograph(). What I didn't really understand is which layer to use for message passing and how the structure of the GNN model should look like. I have only one edge type and would like to have a simple recommneder system for similiarity analysis.
Thats something you need to figure out experimentally. You can start with GCNConv and then try others. :) Simply stack a couple of layers and build a network head for link prediction. That means - use the final node level embeddings, concatenate them according to the edges and predict one value per edge. Good luck!
Collaborative Filtering is such an underrated computation method, imho
Many real life problems can be re-interpteted as a recommendation problem. It's just a matter of perspective, but the results can be huge!
that s an amazing job! clearly honest deep effort i ncompliling such huge info in a compact and unified language! Congrats on this level of sophistication! Predict an even brighter future for you!
Thank you very much for the kind words!
your content and explanation is Incredibly helpful. Thank you
As usual Awesome! Just waiting for credit card fraud detection video so that I can learn to transform a tabular data to graph data
great explanation and overview as always :D
long time no see, i am glad to watch video released by you
Thank you :) how are you? Hope everything is fine
Your clarity solved a bunch of my open questions that arose digging into the topic. Thank you!
Felicitaciones, un trabajo excepcional. Saludos
Thanks great video!
It's very helpful , Thank a lot
I loved it thank you very much
Great job
may i ask could we build a decision support system based on knowledge graphs using graph neural networks?
Sure, I think nothing speaks against trying it :)
Many people use GNNs on knowledge graphs so it might also work for DSS's
Maybe a practical example?
I like your videos so much, so plz keep going on. My question is how can I build a simple model of a recommendation system using graph convolutional network?
how did you find these papers? I'm interested in developing a recommendation system for creating outfits for a person i.e. you add a shirt to your shopping cart, the recommender will show pants, shoes, and belts that match the shirt
Which papers are you referring to? :)
Just did a search on Google scholar
Awesome !!!! I was breaking my head to structure the vast information available on GNN. Love this video and ily bro !
Haha ily2
+
Great video. Thanks !!
Amazing Content. Exceptional
Wonderful explanation!
this is gold
Can you please do a tutorial on edge classification, please? Practical implementation I mean.
I have a video on node classification :) it's pretty much exactly that, only that the head of the network is different. Instead of one output per node, you have one output per possible edge :)
I must say your videos are great please keep up the good work. My question is how can i implement a simple recommender system with pytorch geometric. I have a custom dataset with two nodes: movies and user. Furthermore there is an edge that describes only the relation of both nodes. I want to create a bipartite graph like the one from kipf & welling and do a similarity analysis. I know how to get the data into the right format. But since I am new to the subject, I am having a hard time implementing a simple gnn model. If anyone here could help me I would be very grateful.
Hi! There is a section on bipartite graphs in Pytorch geometric: pytorch-geometric.readthedocs.io/en/latest/notes/batching.html?highlight=Bipartite#bipartite-graphs
Also I have a couple of videos on how to create custom datasets. Let me know if you need further help!
@@DeepFindr Hello, thank you for the quick feedback. I have now watched many of your videos again. I know how to create a bipartite garph with Heterograph(). What I didn't really understand is which layer to use for message passing and how the structure of the GNN model should look like. I have only one edge type and would like to have a simple recommneder system for similiarity analysis.
Thats something you need to figure out experimentally. You can start with GCNConv and then try others. :)
Simply stack a couple of layers and build a network head for link prediction. That means - use the final node level embeddings, concatenate them according to the edges and predict one value per edge. Good luck!
Do you have any links to blog post / examples showcasing these methods?
I've seen this post on medium: medium.com/decathlontechnology/building-a-recommender-system-using-graph-neural-networks-2ee5fc4e706d
@@DeepFindr thanks so much, keep up the good work