The implementation of the submitted paper "Deep Multi-Behaviors Graph Network for Voucher Redemption Rate Prediction" in SIGKDD 2021 Applied Data Science Track.

Overview

DMBGN: Deep Multi-Behaviors Graph Networks for Voucher Redemption Rate Prediction

The implementation of the accepted paper "Deep Multi-Behaviors Graph Networks for Voucher Redemption Rate Prediction" in SIGKDD 2021 Applied Data Science Track.

DMBGN utilizes a User-Behavior Voucher Graph (UVG) to extract complex user-voucher-item relationship and the attention mechanism to capture users' long-term voucher redemption preference. Experiments shows that DMBGN achieves 10%-16% relative AUC improvement over Deep Neural Networks (DNN), and 2% to 4% AUC improvement over Deep Interest Network (DIN).

Benchmark Dataset

A randomly desensitized sampled dataset from one of the large-scaled production dataset from from Lazada (Alibaba Group) is included. The dataset contains three dataframes corresponding users' voucher collection logs, related user behavior logs and related item features, a detailed description can be found in ./data/README.md file.

We hope this dataset could help to facilitate research in the voucher redemption rate prediction field.

DMBGN Performance

Compared Models:

  • LR: Logistic Regression [1], a shallow model.
  • GBDT: Gradient Boosting Decision Tree [2], a tree-based non deep-learning model.
  • DNN: Deep Neural Networks.
  • WDL: Wide and Deep model [3], a widely accepted model in real industrial applications with an additional linear model besides the deep model compared to DNN.
  • DIN: Deep Interest Network [4], an attention-based model in recommendation systems that has been proven successful in Alibaba.

The experimental results on the public sample dataset are as follows:

Model AUC RelaImpr(DNN) RelaImpr(DIN) Logloss
LR 0.7377 -9.22% -14.28% 0.3897
xgBoost 0.7759 5.40% -0.48% 0.3640
DNN 0.7618 0.00% -5.57% 0.3775
WDL 0.7716 3.73% -2.05% 0.3717
DIN 0.7773 5.90% 0.00% 0.3688
DMBGN_AvgPooling 0.7789 6.54% 0.61% 0.3684
DMBGN_Pretrained 0.7804 7.11% 1.14% 0.3680
DMBGN 0.7885 10.20% 4.06% 0.3616

Note that this dataset is a random sample from dataset Region-C and the performance is different as in the submitted paper due to the smaller sample size (especially xgBoost). However, the conclusion from the experiment results is consistent with the submitted paper, where DMBGN achieves 10.20% relative AUC improvement over DNN and 4.6% uplift over DIN.

image info

How To Use

All experiment codes are organized into the DMBGN_SIGKDD21-release.ipynb jupyter notebook including corresponding running logs, detail code implementation of each model (LR, GBDT, DNN, WDL, DIN, DMBGN) can be found in ./models folder.

To run the experiments, simply start a jupyter notebook and run all code cells in the DMBGN_SIGKDD21-release.ipynb file and check the output logs. Alternatively, you can refer to the existing log outputs in the notebook file. (If you encounter "Sorry, something went wrong. Roload?" error message, just click Reload and the notebook will show.)

To use the DMBGN model, please refer to the code implementation in ./models/DMBGN.py.

Minimum Requirement

python: 3.7.1
numpy: 1.19.5
pandas 1.2.1
pandasql 0.7.3
torch: 1.7.1
torch_geometric: 1.6.3
torch: 1.7.1
torch-cluster: 1.5.8
torch-geometric: 1.6.3
torch-scatter: 2.0.5
torch-sparse: 0.6.8
torch-spline-conv: 1.2.0
torchaudio: 0.7.2
torchvision: 0.8.2
deepctr-torch: 0.2.3
pickle: 4.0

What To Do

  • We are currently deploying DMBGN model online for Lazada voucher related business, the online A/B testing performance will be reported soon.
  • More detailed code comments are being added.

Acknowledgment

Our code implementation is developed based on the Deep Interest Network (DIN) codes from the DeepCTR package, with modification to fit DMBGN model architecture and multi-GPU usage.

We thanks the anonymous reviewers for their time and feedback.

Reference

  • [1] H Brendan McMahan, Gary Holt, David Sculley, Michael Young, Dietmar Ebner,Julian Grady, Lan Nie, Todd Phillips, Eugene Davydov, Daniel Golovin, et al.2013. Ad click prediction: a view from the trenches. InProceedings of the 19thACM SIGKDD international conference on Knowledge discovery and data mining.1222–1230.
  • [2] Guolin Ke, Qi Meng, Thomas Finley, Taifeng Wang, Wei Chen, Weidong Ma,Qiwei Ye, and Tie-Yan Liu. 2017. Lightgbm: A highly efficient gradient boostingdecision tree.Advances in neural information processing systems30 (2017), 3146–3154.
  • [3] Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra,Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, RohanAnil, Zakaria Haque, Lichan Hong, Vihan Jain, Xiaobing Liu, and Hemal Shah.2016. Wide & Deep Learning for Recommender Systems.CoRRabs/1606.07792(2016). arXiv:1606.07792 http://arxiv.org/abs/1606.07792 .
  • [4] Guorui Zhou, Xiaoqiang Zhu, Chenru Song, Ying Fan, Han Zhu, Xiao Ma, YanghuiYan, Junqi Jin, Han Li, and Kun Gai. 2018. Deep interest network for click-throughrate prediction. InProceedings of the 24th ACM SIGKDD International Conferenceon Knowledge Discovery & Data Mining. 1059–1068.
Recommender System Papers

Included Conferences: SIGIR 2020, SIGKDD 2020, RecSys 2020, CIKM 2020, AAAI 2021, WSDM 2021, WWW 2021

RUCAIBox 704 Jan 06, 2023
Code for my ORSUM, ACM RecSys 2020, HeroGRAPH: A Heterogeneous Graph Framework for Multi-Target Cross-Domain Recommendation

HeroGRAPH Code for my ORSUM @ RecSys 2020, HeroGRAPH: A Heterogeneous Graph Framework for Multi-Target Cross-Domain Recommendation Paper, workshop pro

Qiang Cui 9 Sep 14, 2022
RetaGNN: Relational Temporal Attentive Graph Neural Networks for Holistic Sequential Recommendation

RetaGNN: Relational Temporal Attentive Graph Neural Networks for Holistic Sequential Recommendation Pytorch based implemention of Relational Temporal

28 Dec 28, 2022
The implementation of the submitted paper "Deep Multi-Behaviors Graph Network for Voucher Redemption Rate Prediction" in SIGKDD 2021 Applied Data Science Track.

DMBGN: Deep Multi-Behaviors Graph Networks for Voucher Redemption Rate Prediction The implementation of the accepted paper "Deep Multi-Behaviors Graph

10 Jul 12, 2022
Cloud-based recommendation system

This project is based on cloud services to create data lake, ETL process, train and deploy learning model to implement a recommendation system.

Yi Ding 1 Feb 02, 2022
ToR[e]cSys is a PyTorch Framework to implement recommendation system algorithms

ToR[e]cSys is a PyTorch Framework to implement recommendation system algorithms, including but not limited to click-through-rate (CTR) prediction, learning-to-ranking (LTR), and Matrix/Tensor Embeddi

LI, Wai Yin 90 Oct 08, 2022
基于个性化推荐的音乐播放系统

MusicPlayer 基于个性化推荐的音乐播放系统 Hi, 这是我在大四的时候做的毕设,现如今将该项目开源。 本项目是基于Python的tkinter和pygame所著。 该项目总体来说,代码比较烂(因为当时水平很菜)。 运行的话安装几个基本库就能跑,只不过里面的数据还没有上传至Github。 先

Cedric Niu 6 Nov 19, 2022
The official implementation of "DGCN: Diversified Recommendation with Graph Convolutional Networks" (WWW '21)

DGCN This is the official implementation of our WWW'21 paper: Yu Zheng, Chen Gao, Liang Chen, Depeng Jin, Yong Li, DGCN: Diversified Recommendation wi

FIB LAB, Tsinghua University 37 Dec 18, 2022
reXmeX is recommender system evaluation metric library.

A general purpose recommender metrics library for fair evaluation.

AstraZeneca 258 Dec 22, 2022
Fast Python Collaborative Filtering for Implicit Feedback Datasets

Implicit Fast Python Collaborative Filtering for Implicit Datasets. This project provides fast Python implementations of several different popular rec

Ben Frederickson 3k Dec 31, 2022
RecList is an open source library providing behavioral, "black-box" testing for recommender systems.

RecList is an open source library providing behavioral, "black-box" testing for recommender systems.

Jacopo Tagliabue 375 Dec 30, 2022
Code for KHGT model, AAAI2021

KHGT Code for KHGT accepted by AAAI2021 Please unzip the data files in Datasets/ first. To run KHGT on Yelp data, use python labcode_yelp.py For Movi

32 Nov 29, 2022
Codes for AAAI'21 paper 'Self-Supervised Hypergraph Convolutional Networks for Session-based Recommendation'

DHCN Codes for AAAI 2021 paper 'Self-Supervised Hypergraph Convolutional Networks for Session-based Recommendation'. Please note that the default link

Xin Xia 124 Dec 14, 2022
Pytorch domain library for recommendation systems

TorchRec (Experimental Release) TorchRec is a PyTorch domain library built to provide common sparsity & parallelism primitives needed for large-scale

Meta Research 1.3k Jan 05, 2023
6002project-rl - An implemention of offline RL on recommender system

An implemention of offline RL on recommender system @author: misajie @update: 20

Tzay Lee 3 May 24, 2022
Codes for CIKM'21 paper 'Self-Supervised Graph Co-Training for Session-based Recommendation'.

COTREC Codes for CIKM'21 paper 'Self-Supervised Graph Co-Training for Session-based Recommendation'. Requirements: Python 3.7, Pytorch 1.6.0 Best Hype

Xin Xia 43 Jan 04, 2023
Movies/TV Recommender

recommender Movies/TV Recommender. Recommends Movies, TV Shows, Actors, Directors, Writers. Setup Create file API_KEY and paste your TMDB API key in i

Aviem Zur 3 Apr 22, 2022
Graph Neural Network based Social Recommendation Model. SIGIR2019.

Basic Information: This code is released for the papers: Le Wu, Peijie Sun, Yanjie Fu, Richang Hong, Xiting Wang and Meng Wang. A Neural Influence Dif

PeijieSun 144 Dec 29, 2022
Jointly Learning Explainable Rules for Recommendation with Knowledge Graph

Jointly Learning Explainable Rules for Recommendation with Knowledge Graph

57 Nov 03, 2022
Recommendation Systems for IBM Watson Studio platform

Recommendation-Systems-for-IBM-Watson-Studio-platform Project Overview In this project, I analyze the interactions that users have with articles on th

Milad Sadat-Mohammadi 1 Jan 21, 2022