Please note that all times are in MST zone.
|1:50pm – 2:00pm||Opening and Prizes||10 min|
|2:00pm – 2:15pm||Task 3: Cross-market recommendation — Invited Talk |
(University of Amsterdam, Google, Spotify)
|2:15pm – 2:30pm||Task 3: 1st prize (Team “OPDAI”)||15 min|
|2:30pm – 2:45 pm||Task 3: 2nd prize (Team “WSDM_Coggle_”)||15 min|
|2:45pm – 3:00pm||Task 3: 3rd prize (Team “BiuG”)||15 min|
|3:00pm – 3:30pm||Coffee break||30 min|
|3:30pm – 3:45pm||Task 2: Temporal link prediction — Invited Talk (Intel, Amazon)||15 min|
|3:45pm – 4:00pm||Task 2: 1st prize (Team “AntGraph”)||15 min|
|4:00pm – 4:15pm||Task 2: 2nd prize (Team “nothing here”)||15 min|
|4:15pm – 4:30pm||Task 2: 3rd prize (Team “NodeInGraph”)||15 min|
|4:30pm – 4:45pm||Task 1: User retention score prediction — Invited Talk (iQIYI)||15 min|
|4:45pm – 5:00pm||Task 1: 1st prize (Team “QDU”)||15 min|
|5:00pm – 5:15pm||Task 1: 2nd prize (Team “ASDF”)||15 min|
|5:15pm – 5:30pm||Task 1: 3rd prize (Team “二次元的死肥宅”|
(Otaku and ACG Fans))
|5:30pm – 5:40pm||Closing||10 min|
Task sponsor: iQIYI
Task abstract: iQIYI is the world’s leading movie and video streaming platform, with nearly 8 billion hours spent on iQIYI app each month and over 500 million monthly active users. It features highly popular original content, as well as a comprehensive library of other professionally-produced content and user-generated content. iQIYI app uses deep-learning AI algorithms and massive user data, to produce content that caters to various user tastes, and to deliver superior entertainment experience to them. User retention is a key indicator to measure the users’ satisfaction. It uses a retention score for the next N days to evaluate the retention. For example, a user having retention score 3 for the next 7 days means this user would launch the iQIYI app in 3 separate days during the period. It is very challenging to predict the retention score. It not only depends on the individual, but also on the trends of the entertainment, e.g., the number of daily users fluctuates severely when some blockbuster original contents are released. The task’s goal is to predict user retention scores using deep models.
Task sponsor: Intel / Amazon
Task abstract: Real world datasets can often be expressed as graphs, with entities as nodes and interactions as edges. Examples include user-user interaction in social networks, user-item interactions in recommender systems, etc. Moreover, these graphs are often in practice temporal, with new edges coming in with timestamps. Contrary to link prediction which asks if an edge exists between two nodes on a partially observed graph, temporal link prediction asks if an edge will exist between two nodes within a given time span. It is more useful than traditional link prediction as one can then build multiple applications around the model, such as forecasting the demand of customers in E-commerce, or forecasting what event will happen in a social network, etc. We are expecting a single model that can handle two kinds of data simultaneously: a dynamic event graph with entities as nodes and events as edges, and a user-item graph with users and items as nodes and various interactions as edges. The task will be predicting whether an edge of a given type will exist between two given nodes before a given timestamp.
Task sponsor: University of Amsterdam / Google / Spotify
Task abstract: E-commerce companies often operate across markets; for instance Amazon has expanded their operations and sales to 18 markets (i.e. countries) around the globe. The cross-market recommendation concerns the problem of recommending relevant products to users in a target market (e.g., a resource-scarce market) by leveraging data from similar high-resource markets, e.g. using data from the U.S. market to improve recommendations in a target market. The key challenge, however, is that data, such as user interaction data with products (clicks, purchases, reviews), convey certain biases of the individual markets. Therefore, the algorithms trained on a source market are not necessarily effective in a different target market. Despite its significance, small progress has been made in cross-market recommendation, mainly due to a lack of experimental data for the researchers. In this WSDM Cup challenge, we provide user purchase and rating data on various markets, enriched with review data in different languages, with a considerable number of shared item subsets. The goal is to improve individual recommendation systems in these target markets by leveraging data from similar auxiliary markets.
Title: Introduction to User Retention Prediction and Motivation (Task 1)
Abstract: User retention is a key indicator to measure the users’ satisfaction. A well-defined metric and accurate prediction can help user growth and causal inference. In this talk, we present the motivation, real-world datasets and our baseline solution for the task.
Speaker Bio: Zane Zhang is a senior manager of Business Intelligence Platforms at iQIYI, where he leads the video recommendation and user growth teams. The teams drive business strategies through data insights and solve their users’ problems using machine learning stack and big data services. Zane has been in the data field for the past 12 years and his experience spans across machine learning, computational statistics, and its domain-specific applications. He holds a MS in Computer Science from Fudan University.
Title: Temporal Link Prediction Challenge – Background, Results, Lessons and Future Plans (Intel / Amazon) (Task 2)
Abstract: This talk will cover the following aspects of the Temporal Link Prediction Challenge for WSDM Cup 2022: (1) Our motivation of hosting this challenge, with a review of previous works in temporal link prediction. (2) How we build the two datasets. (3) The results and lessons learned from the challenge. (4) Our future plans on establishing a new leaderboard.
Speaker Bio: Quan Gan is an Applied Scientist at Amazon Shanghai AI Lab. He is currently one of the main developers of the Deep Graph Library (DGL). His research interest is primarily in graph neural networks on heterogeneous or dynamic graphs, large-scale training, and system optimization.
Title: XMRec: Cross-market Recommendation (University of Amsterdam / Google / Spotify) (Task 3)
Abstract: E-commerce companies often operate across markets; for instance Amazon has expanded their operations and sales to 18 markets (i.e. countries) around the globe. The cross-market recommendation concerns the problem of recommending relevant products to users in a target market (e.g., a resource-scarce market) by leveraging data from similar high-resource markets, e.g. using data from the U.S. market to improve recommendations in a target market. The key challenge, however, is that data, such as user interaction data with products (clicks, purchases, reviews), convey certain biases of the individual markets. Therefore, the algorithms trained on a source market are not necessarily effective in a different target market. Despite its significance, small progress has been made in cross-market recommendation, mainly due to a lack of experimental data for the researchers. In this WSDM Cup challenge, we provide user purchase and rating data on various markets with a considerable number of shared item subsets. The goal is to improve individual recommendation systems in these target markets by leveraging data from similar auxiliary markets. In this talk, we will provide further details about the task setup, shared data, and the participation.
Speaker Bio: Mohammad Aliannejadi is an assistant professor at IRLab, University of Amsterdam. Before that, Mohammad obtained his Ph.D. from USI, Switzerland, where he worked on novel approaches of information access in conversational IR. His research interests are conversational search and recommendation, user modeling and simulation, as well as cross-market and cross-lingual recommendation. Mohammad has been an active member of the community, organizing several data challenges and workshops on various topics, including the XMRec Workshop at RecSys 2021, IGLU Contest at NeurIPS 2021, ConvAI at EMNLP 2020, and TREC CAsT 2022.
WSDM Cup 2022 Prize Winner Reports
Task 1 (User Retention Prediction)
- An Effective Ensemble Framework with Multichannel Time Series for User Retention Prediction [pdf]
Zhiruou Li, Zhihui Cui, Shunyao Wu (Team “QDU”)
- Retention Score Prediction via TabNet [pdf]
Qian Zhang, Huayuan Sun (Team “ASDF”)
- Dynamic RNN: An Effective Approach for User Retention Prediction [pdf]
Chenfei Kang, Wei Wang, Weiqiang Tang, Sibei Li, Sihan Lu (Team “Otaku & ACG Fans”)
Task 2 (Temporal Link Prediction)
- An Effective Graph Learning based Approach for Temporal Link Prediction [pdf]
Qian Zhao, Shuo Yang, Binbin Hu, Zhiqiang Zhang, Yakun Wang, Yusong Chen, Jun Zhou, Chuan Shi (Team “AntGraph”)
- HTGN-BTW: Heterogeneous Temporal Graph Network with Bi-Time-Window Training Strategy for Temporal Link Prediction [pdf]
Chongjian Yue, Lun Du, Qiang Fu, Wendong Bi, Hengyu Liu, Yu Gu, Di Yao (Team “nothing here”)
- Temporal Link Prediction with Network Embedding [pdf]
Manting Shen (Team “NodeInGraph”)
Task 3 (Cross Market Recommendation)
- An Effective Way for Cross-Market Recommendation with Hybrid Pre-Ranking and Ranking Models [pdf]
Qi Zhang, Zijian Yang, Yilun Huang, Jiarong He, Lixiang Wang (Team “OPDAI”)
- A Practical Two-stage Ranking Framework for Cross-market Recommendation [pdf]
Zeyuan Chen, He Wang, Xiangyu Zhu, Haiyan Wu, Congcong Gu, Shumeng Liu, Jinchao Huang, Wei Zhang (Team “WSDM_Coggle_”)
- A Learning to Rank method for Cross-market Recommendation [pdf]
Peng Zhang, Rongxiu Gao, Changyv Li, Baoyin Liu, Sheng Zhou (Team “BiuG”)