HYUIDSL/SVD_FMs

Code for Simple and effective recommendations using implicit feedback-aware factorization machines

Python

Code for Simple and effective recommendations using implicit feedback-aware factorization machines

0. Overview

Our paper mainly proposes SVD embedding for factorization machines (FM/DeepFM)
Followed Most of the DeepFM/FM code structure from https://github.com/rixwew/pytorch-fm

1. Basic Requirements.

Code Tested on cuda 11.8 and python 3.10.11~12
pip Install requirements.txt

2. How to Use

run new_test.py

To try SVD embedding, make args.embedding_type="SVD"
To change model options(FM/DeepFM) change args.embedding_type
Various hyperparamters to test can be changed based on your preference.
datasets: other more datasets were used but here we upload ml100k and ml1m

3. Algorithms

SVD Embedding applied FM/DeepFM can be seen in model/SVD_emb folder
How we implemented negative sampling can be seen in util/negativesampler.py
Our frequency-based negative sampling (abbreviated FNS in paper, ) can be used by making args.isuniform=False