About ^top

The RecSys Challenge 2022 is organized by Nick Landia (Dressipi), Bruce Ferwerda (Jönköping University, Sweden), Saikishore Kalloori (ETH Zürich, Switzerland), and Abhishek Srivastava (IIM Visakhapatnam, India).

Dressipi ^top

Dressipi are the fashion-AI experts, providing product and outfit recommendations to leading global retailers.

Our recommendations enable retailers to create new product discovery experiences that are personalized and inspiring and can be used at all steps of the shopper journey.

Our algorithms enable retailers to make better buying and merchandising decisions by more accurately forecasting product demand and size ratios.

Our focus is to provide the world’s best apparel recommendations and predictions. We do this by taking a domain specific approach across the data we collect and create, how we structure that data and the models we build. Everything we do is optimized to handle the nuances of fashion.

We work with brands across the US, UK, Europe, and Australia and outperform every competitor when A/B tested.

Challenge Task ^top

This year’s challenge focuses on fashion recommendations. When given user sessions, purchase data and content data about items, can you accurately predict which fashion item will be bought at the end of the session?

The content data consists of descriptive labels of the items (such as color, length, neckline, sleeve style, etc.). The labels have been assigned using Dressipi’s human-in-the-loop system where fashion experts review, correct and confirm the correctness of the labels, so we expect this to be a dataset of high accuracy and quality.

Fig 1: Example Session and Purchase Data

It’s important to be able to make recommendations that respond to what the user is doing during the current session to create the best experience possible that results in a purchase. Nuances of the fashion domain make accurate in-session predictions more critical than in other domains:

• On average 51% of total visitors are new (Dressipi Data) which means there is no historical data available and we solely have to rely on current session activity.
• Even for the other half of visitors that have historical data, trends and other external factors change user preferences much more quickly than in other domains, meaning the historical data might no longer be representative of the user’s interests on a case-to-case basis. This places even more importance on having a highly accurate in-session recommender that can be pulled into the mix.
• Sessions can be pretty short so we need to be able to make accurate predictions as early as possible, before the user bounces.

Dataset ^top

As part of this challenge, Dressipi will be releasing a public dataset of 1.1 million online retail sessions that resulted in a purchase. In addition, all items in the dataset have been labeled with content data and the labels are supplied. We refer to the label data as item features. The dataset is sampled and anonymized.

• Sessions: The items that were viewed in a session. In this dataset a session is equal to a day, so a session is one user's activity on one day.
• Purchases: The purchase that happened at the end of the session. One purchased item per session.
• Item features: The label data of items. Things like “color: green,” “neckline: v-neck,” etc.

The image below is an illustration of what the content data could look like for a given dress (this is a made-up example). In the dataset the label data has been anonymised by using ids: you will not get the cleartext labels like “neckline: v-neck” but rather ids representing the same data.

A more detailed description can be found by clicking on the button below.

Fig 2: Example Content Data

Rules ^top

• Anyone (18+) can participate
• Do not use external data feeds
• Do not use models trained on external data
• When you train the model, only use the data from the training file. Do not use any data from the "leaderboard" or "final" test files
• When predicting, treat each test session independently of all other test sessions (i.e., when predicting for test session B, the model should not have any knowledge of test session A. Even if that came before it in terms of time-stamp)

Accepted contributions will be presented during the RecSys Challenge Workshop in 2022.

Participation and Data ^top

The data for this year's challenge is provided by Dressipi.

Consult the Google Groups if you are experiencing any issues.

The dataset and detailed information on the challenge participation will be provided after creating an account.

Participation for the challenge is subject to the acceptance of the Terms & Conditions.

Paper Submission Guidelines ^top

Submission website: EasyChair

• All participants of the challenge are invited to submit if they consider their submission particularly effective, novel, otherwise interesting, or exploiting identified particularities of the data.
• Note: paper submission is mandatory if you want to be eligible for a prize. Accepted papers are given a presentation slot at the workshop. At least one author of each accepted paper must attend the workshop and present their work. Please note that a badly written paper or absence of presence at workshop, may prevent you from being eligible for the prize. Please contact the workshop organization if none of the authors will be able to attend the workshop.
• Page limit: 7 pages + references (ACM SIG Format) Note: This is in reference to the latest single-column ACM template.
• Anonymization of submissions is not required; please include your team name in abstract and text, as well as a link to your code repository, the achieved score, and a reference to the RecSys Challenge Website (http://www.recsyschallenge.com/2022/). Note: This will be replaced with a reference to an overview paper in the RecSys proceedings for the camera-ready version.
• The submitted papers will be evaluated based on novelty, clarity, and presented empirical results.
• Each paper will be reviewed by at least three PC members.
• Our proceedings will be published in the ACM Digital Library within its International Conference Proceedings Series.
• Accepted papers must be presented in the RecSys Challenge Workshop.

Workshop Program and Accepted Papers ^top

The RecSys Challenge Workshop will take place on September 22nd, 2022
All times are Pacific Daylight Time

Time	Session
14:00 - 14:12	Opening
14:15 - 14:27	Lightweight Model for Session-Based Recommender Systems with Seasonality Information in the Fashion Domain Nicola Della Volpe, Lorenzo Mainetti, Alessio Martignetti, Andrea Menta, Riccardo Pala, Giacomo Polvanesi, Francesco Sammarco, Fernando Benjamín Pérez Maurera, Cesare Bernardis and Maurizio Ferrari Dacrema
14:30 - 14:42	United we stand, divided we fall: leveraging ensembles of recommenders to compete with budget constrained resources Pietro Maldini, Alessandro Sanvito and Mattia Surricchio (virtual)
14:45 - 14:57	Session-Based Recommendation by combining Probabilistic Models and LSTM Costas Panagiotakis and Harris Papadakis (virtual)
15:00 - 15:12	SIHG4SR: Side Information Heterogeneous Graph for Session Recommender Chendi Xue, Xinyao Wang, Yu Zhou, Ke Ding, Jian Zhang, Rita Brugarolas Brufau and Eric Anderson (virtual)
15:15 - 15:27	Fashion Recommendation with a real Recommender System Flow Qi Zhang, Guohao Cai, Wei Guo, Yi Han, Zhenhua Dong, Ruiming Tang and Liangbi Li (virtual)
15:30 - 16:00	Coffee Break
16:00 - 16:12	Simple and Efficient Recommendation Strategy for Warm/Cold Sessions for RecSys Challenge 2022 Hyunsung Lee, Sungwook Yoo, Andrew Yang, Wonjun Jang and Chiwan Park
16:15 - 16:27	LightGBM using Enhanced and De-biased Item Representation for Better Session-based Fashion Recommender Systems Jiangwei Luo, Wenxuan Zhao, Ye Tang, Zhou Zhou, Huimin Xiong and Zhulin Tao (virtual)
16:30 - 16:42	A Diverse Models Ensemble for Fashion Session-Based Recommendation Benedikt Schifferer, Jiwei Liu, Sara Rabhi, Gilberto Titericz, Chris Deotte, Gabriel de Souza P. Moreira, Ronay Ak and Kazuki Onodera
16:45 - 16:57	Session-based Recommendation with Transformer Yichao Lu, Zhaolin Gao, Zhaoyue Cheng, Jianing Sun, Bradley Brown, Guangwei Yu, Anson Wong, Felipe Pérez and Maksims Volkovs
17:00 - 17:12	Industrial Solution in Fashion-domain Recommendation by an Efficient Pipeline using GNN and Lightgbm Zzh, Wei Zhang and Tao Wen (virtual)
17:15 - 17:30	Closing remarks

Organization ^top

Program Committee ^top

• Vito Walter Anelli, Politecnico di Bari
• Luca Belli, Twitter Cortex
• Alejandro Bellogin, Universidad Autonoma de Madrid
• Ludovico Boratto, University of Cagliari
• Ludovik Coba, Koa Health
• Tommaso Di Noia, Politecnico di Bari
• Dietmar Jannach, University of Klagenfurt
• Marko Tkalcic, University of Primorska
• Wolfgang Wörndl, Technical University of Munich

Organizers ^top

• Nick Landia, Dressipi
• Bruce Ferwerda, Jönköping University
• Saikishore Kalloori, ETH Zürich
• Abhishek Srivastava, Indian Institute of Management, Visakhapatnam
• Frederick Cheung, Dressipi
• Donna North, Dressipi

Advisor(s) ^top

• Vito Walter Anelli, Politecnico di Bari