The RecSys 2023 Challenge will be organized by Sarang Brahme, Rahul Agarwal (ShareChat), Abhishek Srivastava (IIM Visakhapatnam, India), Liu Yong (Huawei, Singapore) and Athirai Irissappane (Amazon, USA) based on the data provided by ShareChat. This year’s challenge will focus on online advertising, improving deep funnel optimization, and user privacy.
The challenge is brought to you by ShareChat. ShareChat is India’s largest homegrown social media company, with 400+ million MAUs across all its platforms. Headquartered in Bengaluru, ShareChat is spreading its team globally across India, the USA, and Europe. We have the best-in-class AI & ML technology and the strongest feed ranking system powering our growth. We aim to create a million monetizable creators with USD 450 million in creator earnings across ShareChat and Moj by 2025.
Online advertising has been a multi-billion dollar industry since the early 2000 and has played a significant role in the growth of the internet. The key advantage of online advertising over conventional mass advertising is its inherent ability to personalize to users, democratizing advertising and enabling businesses of all sizes to participate, and providing the measurable impact of money spent to the advertisers. Over the past two decades, the nature of online advertising has also evolved tremendously from pure banner-based advertising, where advertisers were charged based on the number of ad impressions, to deep funnel optimizations, where advertisers can optimize for eventual sales.
The efficacy of deep funnel optimization required extensive personalization and opened up rich problems in real-time auction design, large-scale machine learning, modeling delayed feedback, and behavioral understanding. As these systems matured, we also started developing a rich understanding of the need to preserve user privacy, ensure AI fairness, and prevent adversarial exploitation of the platform. In this challenge, we aim to provide a real-world ad dataset from the Sharechat and Moj apps to act as a benchmark for research into deep funnel optimization with a focus on user privacy
The dataset corresponds to roughly 10M random users who visited the ShareChat + Moj app over three months. We have sampled each user's activity to generate 10 impressions corresponding to each user. Our target variable is whether there was an install for an app by the user or not.
To represent a user, several features are provided:
We also have features corresponding to ads
To capture the historical interactions between users and ads, we also provide
Every row of the data has an associated numeric id and represents an ad impression shown to the user and whether it resulted in a click on the ad and subsequently an install or not.
We do not provide the semantics of the individual features.
The training data consists of subsampled impressions/clicks/installs from the past 2 weeks and aims to predict the probability of install for the 15th day.
|27 March, 2023||
Start RecSys Challenge
|11 Apr, 2023||Submission System Open|
|13 Apr, 2023||Leaderboard live|
|22nd June, 2023
||End RecSys Challenge|
|28th June, 2023
Final Leaderboard & Winners
EasyChair open for submissions
|3rd July, 2023
Upload code of the final predictions
|14 July, 2023||Paper Submission Due|
|1 August, 2023||Paper Acceptance Notifications|
|14 August, 2023||Camera-Ready Papers|
|Sept 3rd week||RecSys Challenge Workshop|
Submission website: EasyChair
|9:15-9:30||A Simple and Robust Ensemble For Click-Through Rate Prediction Xingmei Wang and Yankai Wanga|
|9:30-9:45||Predicting Conversion Rate in Advertising Systems: A Two-Stage Approach with LightGBM Lulu Wang, Yu Zhang, Huayang Zhao, Zhewei Song and Jiaxin Hu|
|9:45-10.00||Integrating Explicit and Implicit Feature Interactions for Online Ad Installation Forecasting Jiawei Jiang, Bing Wang and Jingyuan Wang|
|10:00-10:15||Capturing Performance and Privacy by Assembling Avengers of Online Advertising Taehee Kim, Seungyun Baek, Taehyeon Jeon, Hojin Jung, Joonhong Kim and Taeho Lee|
|10:15-10:30||Lightweight Boosting Models for User Response Prediction Using Adversarial Validation Hyeonwoo Kim and Wonsung Lee|
|11:15-11:30||Robust User Engagement Modeling With Transformers and Self Supervision Yichao Lu and Maksims Volkovs|
|11:30-11:45||Pessimistic Rescaling and Distribution Shift of Boosting Models for Impression-Aware Online Advertising Recommendation Paolo Basso, Arturo Benedetti, Nicola Cecere, Alessandro Maranelli, Salvatore Marragony, Samuele Peri, Andrea Riboni, Alessandro Verosimile, Davide Zanutto and Maurizio Ferrari Dacrem|
|11:45-12:00||Graph Enhanced Feature Engineering for Privacy Preserving Recommendation Systems Chendi Xue, Xinyao Wang, Yu Zhou, Poovaiah Palangappa, Rita Brugarolas Brufau, Aasavari Dhananjay Kakne, Ravi Motwani, Ke Ding and Jian Zhang|
|12:00-12:15||A Simple yet Strong Approach for Installation Prediction in ShareChat Ads Xiaoteng Shen, Liangcai Su, Zhutian Lin and Xiao Xi|
|12:15-12:30||Large Scale CVR Prediction through Hierarchical History Modeling Qi Zhang, Zhibin Zhang, Biao Lu, Bangzheng He and Liangbi Li|