search

Guanyu-Lin / MAN

MIT License
12 stars 0 forks source link

what about three demains? #1

Open hallucination11 opened 9 months ago

hallucination11 commented 9 months ago

how to design shared embedding matrix with three demains ? -.-

Guanyu-Lin commented 9 months ago

Thanks for your question! you can modify "reco_utils/recommender/deeprec/models/sequential/sequential_base_model.py" to add another domain.

hallucination11 commented 9 months ago

Thanks for your answer! I mean if i have 3 domains, whether i need create three shared embedding matrix M ? what about 4 or 5 domains? Or this work is just aim to 2 domains ?

Guanyu-Lin commented 9 months ago

You only need to create one shared (global) embedding M and three independent embeddings for each domain. Our motivation is that each domain has its specific features and shared knowledge across all of them.

hallucination11 commented 9 months ago

Thanks !

MuMuBai commented 8 months ago

Is there a PyTorch version of the code available?

Guanyu-Lin commented 8 months ago

Is there a PyTorch version of the code available?

Sorry, we only have the Tensorflow version as most industrial recommender systems are based on TensorFlow. If you want to implement a Pytorch version, we would appreciate it and are willing to provide any assistance.