sandy4321/beta-recsys

Beta-RecSys: Build, Evaluate and Tune Automated Recommender Systems

|Installation | Quick Start | Documentation | Contributing | Getting help | Citation|

Beta-RecSys an open source project for Building, Evaluating and Tuning Automated Recommender Systems. Beta-RecSys aims to provide a practical data toolkit for building end-to-end recommendation systems in a standardized way. It provided means for dataset preparation and splitting using common strategies, a generalized model engine for implementing recommender models using Pytorch with a lot of models available out-of-the-box, as well as a unified training, validation, tuning and testing pipeline. Furthermore, Beta-RecSys is designed to be both modular and extensible, enabling new models to be quickly added to the framework. It is deployable in a wide range of environments via pre-built docker containers and supports distributed parameter tuning using Ray.

Installation

conda

If you use conda, you can install it with:

conda install beta-rec

pip

If you use pip, you can install it with:

pip install beta-rec

Docker

We also provide docker image for you to run this project on any platform. You can use the image with:

1. Pull image from Docker Hub

   docker pull betarecsys/beta-recsys:latest
   

2. Start a docker container with this image (Make sure the port 8888 is available on you local machine, or you can change the port in the command)

   docker run -ti --name beta-recsys -p 8888:8888 -d beta-recsys
   

3. Open Jupyter on a brower with this URL:

   http://localhost:8888
   

4. Enter root as the password for the notebook.

Quick Start

Downloading and Splitting Datasets

from beta_rec.datasets.movielens import Movielens_100k
from beta_rec.data import BaseData
dataset = Movielens_100k()
split_dataset = dataset.load_leave_one_out(n_test=1)
data =  BaseData(split_dataset)

Training model with MatrixFactorization

config = {
    "config_file":"./configs/mf_default.json"
}
from beta_rec.recommenders import MatrixFactorization
model = MatrixFactorization(config)
model.train(data)
result = model.test(data.test[0])

where a default config josn file ./configs/mf_default.json will be loaded for traning the model.

Tuning Model Hyper-parameters

config = {
    "config_file":"../configs/mf_default.json",
    "tune":True,
}
tune_result = model.train(data)

Experiment with multiple models

from beta_rec.recommenders import MatrixFactorization
from beta_rec.experiment.experiment import Experiment

# Initialise recommenders with their default configuration file

config = {
    "config_file":"configs/mf_default.json"
}

mf_1 = MatrixFactorization(config)
mf_2 = MatrixFactorization(config)

# Run experiments of the recommenders on the selected dataset

Experiment(
  datasets=[data],
  models=[mf_1, mf_2],
).run()

where the model will tune the hyper-parameters according to the specifed tuning scheme (e.g. the default for MF).

Models

The following is a list of recommender models currently available in the repository, or to be implemented soon.

General Models

Model	Paper	Colab
MF	Neural Collaborative Filtering vs. Matrix Factorization Revisited, arXiv’ 2020
GMF	Generalized Matrix Factorization, in Neural Collaborative Filtering, WWW 2017
MLP	Multi-Layer Perceptron, in Neural Collaborative Filtering, WWW 2017
NCF	Neural Collaborative Filtering, WWW 2017
CMN	Collaborative memory network for recommendation systems, SIGIR 2018
NGCF	Neural graph collaborative filtering, SIGIR 2019
LightGCN	LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation, SIGIR 2020
LCF	Graph Convolutional Network for Recommendation with Low-pass Collaborative Filters
VAECF	Variational autoencoders for collaborative filtering, WWW 2018
UserKNN	User-based K-Nearest Neighbour Recommender in Factorization Meets the Neighborhood: a Multifaceted Collaborative Filtering, KDD 2008
ItemKNN	Item-based K-Nearest Neighbour Recommender in Item-Based Collaborative Filtering Recommendation, WWW 2001
MixGCF	MixGCF: An Improved Training Method for Graph Neural Network-based Recommender Systems, KDD 2021
UltraGCN	UltraGCN: Ultra Simplification of Graph Convolutional Networks for Recommendation, CIKM 2021
SGL	Self-supervised Graph Learning for Recommendation, SIGIR 2021
SimGCL	Are Graph Augmentations Necessary? Simple Graph Contrastive Learning for Recommendation, SIGIR 2022

Sequential Models

Model	Paper	Colab
NARM	Neural Attentive Session-based Recommendation, CIKM 2017
Caser	Personalized Top-N Sequential Recommendation via Convolutional Sequence Embedding, WSDM 2018
GRU4Rec	Session-based recommendations with recurrent neural networks, ICLR 2016
SasRec	Self-attentive sequential recommendation. ICDM 2018
MARank	Multi-Order Attentive Ranking Model for Sequential Recommendation, AAAI 2019
NextItnet	A Simple Convolutional Generative Network for Next Item Recommendation, WSDM 2019
BERT4Rec	BERT4Rec: Sequential recommendation with bidirectional encoder representations from transformer, CIKM 2019
TiSASRec	Time Interval Aware Self-Attention for Sequential Recommendation. WSDM'20

Recommendation Models with Auxiliary information

Baskets/Sessions

Model	Paper	Colab
Triple2vec	Representing and recommending shopping baskets with complementarity, compatibility and loyalty, CIKM 2018
VBCAR	Variational Bayesian Context-aware Representation for Grocery Recommendation, arXiv’ 2019
TVBR	Temporal Variational Bayesian Representation Learning for Grocery Recommendation

Knowledge Graph

• KGAT: Kgat: Knowledge graph attention network for recommendation. SIGKDD 2019

If you want your model to be implemented by our maintenance team (or by yourself), please submit an issue following our community instruction.

Recent Changing Logs ---> See version release.

Contributing

This project welcomes contributions and suggestions. Please make sure to read the Contributing Guide before creating a pull request.

Community meeting

• Meeting time: Saturday (1:30 – 2:30pm UTC +0) / (9:30 – 10:30pm UTC +8)
• Meeting minutes: notes
• Meeting recordings: [recording links]: Can be found in each meeting note.

Discussion channels

• Slack:
• Mailing list: TBC

Citation

If you use Beta-RecSys in you research, we would appreciate citations to the following paper:

@inproceedings{meng2020beta,
  title={BETA-Rec: Build, Evaluate and Tune Automated Recommender Systems},
  author={Meng, Zaiqiao and McCreadie, Richard and Macdonald, Craig and Ounis, Iadh and Liu, Siwei and Wu, Yaxiong and Wang, Xi and Liang, Shangsong and Liang, Yucheng and Zeng, Guangtao and others},
  booktitle={Fourteenth ACM Conference on Recommender Systems},
  pages={588--590},
  year={2020}
}