Decoding Online-Shopping Habits to Enable Smart Recommendations: Amazon KDD Cup 2023 Challenge

Introduction

The world of e-commerce is transforming. Now more than ever, understanding customer shopping intentions is the key to unlocking personalized shopping experiences. Imagine a scenario where an online store accurately predicts what you’re searching for, and recommends the perfect product just for you. This is the power of session-based recommendation, an increasingly popular approach that harnesses session data to predict next purchase.

Let’s dive into the triumphs of the Amazon KDD Cup 2023 and meet the exceptional minds behind this transformative AI endeavour.

In the dynamic realm of data mining and machine learning, session-based recommendation has garnered immense interest. However, its real-world application under multilingual and imbalanced conditions has remained largely unexplored. That’s where the “Multilingual Shopping Session Dataset” comes into play, a vast repository of millions of user sessions from six diverse locales, boasting major languages like English, German, Japanese, French, Italian, and Spanish. The dataset presents a unique challenge due to the imbalanced distribution of products across languages, with fewer instances of French, Italian, and Spanish products compared to English, German, and Japanese.

From Shopping Sessions to Smart Recommendations

The Amazon KDD Cup 2023 competition revolved around three compelling tasks, each designed to predict what a user is likely to buy next based on their recent shopping history:

Task 1: Next Product Recommendation
In this task, participants aimed to predict the product that a customer is most likely to purchase next from a list of products they have already viewed. The challenge featured three languages — Japanese, English, and German — with relatively more data in the train set.

Task 2: Next Product Recommendation for Underrepresented Languages
Task 2 shared similarities with Task 1 but introduced a twist. The dataset included French, Spanish, and Italian languages, but this time with limited training data. Participants were encouraged to employ transfer learning to enhance their models’ performance on the test set.

Task 3: Next Product Title Generation
Task 3 presented an intriguing challenge: generating the title of a product that a customer desires to buy next, even if that product is not listed in the provided data. This task called for innovative approaches to cater to the diverse needs of shoppers.

The Multilingual Shopping Session Dataset

To facilitate the competition, Amazon released the Multilingual Shopping Session Dataset, a treasure trove of anonymised customer sessions containing products from six locales: English, German, Japanese, French, Italian, and Spanish. The dataset included user sessions — a chronological list of products a user engaged with — and product attributes such as title, price, brand, color, and description.

The dataset was thoughtfully divided into three splits: train, phase-1 test, and phase-2 test. Task 1 and Task 2 train sets had approximately 10 times more data for each language compared to the test sets. For Task 3, the test set featured products not found in the training set, challenging participants to generate accurate product titles based on user sessions.

Measuring Success: Metrics and Evaluation

The competition employed Mean Reciprocal Rank (MRR) as the metric to evaluate the success of Task 1 and Task 2. For Task 3, the bilingual evaluation understudy (BLEU) metric was utilized. These metrics ensured fair and accurate evaluation, and the competition organisers made the dataset publicly available, enabling further research and improvement in the field of recommender systems.

Meet the Exceptional Winners

Now, let’s meet the brilliant minds whose innovations triumphed in the Amazon KDD Cup 2023 competition:

NVIDIA-Merlin

Chris Deotte
With a diverse background in mathematics, graphic artistry, photography, carpentry, and teaching, I’m now a senior data scientist at Nvidia and a quadruple Kaggle Grandmaster. AI competitions have become my passion, driving me to continuously push the boundaries of innovation.

Benedikt Schifferer
As a manager of a deep learning team at NVIDIA, I bring a wealth of knowledge from developing recommender systems for a German ecommerce company and consulting at McKinsey. The Amazon KDD Cup 2023 has been an incredible platform to showcase our skills and compete with the best.

Jean-Francois Puget
A currency director at NVIDIA, I lead a team of Kaggle competition grandmasters. Participating in ML competitions has become my passion, and I’ve published over 80 scientific papers. Success in ML challenges has driven my focus from publishing to demonstrating the value of my ideas.

Kazuki Onodera
As a Senior Deep Learning Data Scientist at NVIDIA and a Kaggle Competition Grandmaster, I’m dedicated to exploring innovative approaches in the Recommender Systems domain. The Amazon KDD Cup 2023 has been an exciting opportunity to compete on an international stage.

Gilberto Titericz
As a Senior Data Scientist at NVIDIA, I’ve worked with companies like Airbnb, Petrobras, and Siemens. Splitting my time between video games and machine learning, the Amazon KDD Cup 2023 has been a thrilling experience, and I’m honoured to be part of the winning team.

MGTV-REC

He Zhouzhou
I work at Mgtv (Hunantv.com Interactive Entertainment Media Co., Ltd.) and won the RecSys Challenge 2022. The Amazon KDD Cup 2023 was my second competition experience, and it provided valuable insights into AI engineering for recommendation systems.

Wentao Tang
I’m a member of the MGTV team, immersed in the world of recommender systems for the past two years. Participating in the Amazon KDD Cup 2023 has been an honour, and I’ve cherished the chance to compete with exceptional players from around the globe.

Ye Tang
As part of the AI Department in MGTV Corporation, my interests lie in machine learning algorithms for Recommendation Systems, Computer Vision, and Natural Language Processing. The KDD Cup 2023 Challenge provided an exciting opportunity to explore creative approaches in recommendation systems.

Jiangwei Luo
As a senior algorithm engineer with MGTV Corporation, my research interests lie in recommendation algorithm research and development. Participating in the Amazon KDD Cup 2023 has been a thrilling experience, showcasing my skills on an international stage.

ustc-gobble

Chenwang Wu
I’m pursuing a Ph.D. degree at the University of Science and Technology of China, focusing on secure recommender systems and trustworthy machine learning. The Amazon KDD Cup 2023 has been a thrilling journey, competing with exceptional players and gaining invaluable experience.

Leyan Deng
I’m a Ph.D. candidate at the University of Science and Technology of China, with research interests in anomaly detection and spatial-temporal data mining. Participating in the KDD Cup 2023 Challenge has been an enlightening experience, exploring challenging datasets and learning from the brightest minds.

Zhihao Zhu
As a Ph.D. candidate at the University of Science and Technology of China, my research focuses on secure graph neural networks and recommender systems. The Amazon KDD Cup 2023 has been a rewarding experience, challenging me to push the boundaries of recommender systems.

Defu Lian
I’m a professor at the School of Computer Science and Technology, University of Science and Technology of China. My research revolves around spatial data mining, recommender systems, and learning to hash. Participating in the Amazon KDD Cup 2023 has been a remarkable experience, showcasing our research and skills on an international stage.

unirec

Yuxuan Lei (Fire in AIcrowd)
I’m Yuxuan Lei, a Ph.D. candidate at the University of Science and Technology of China. Participating in the Amazon KDD Cup 2023 has been an enriching experience, offering valuable insights into engineering recommender systems and inspiring my future research endeavours.

Xiaolong Chen (CXL in AIcrowd)
I’m a first-year graduate student at the University of Science and Technology of China, focusing on recommendation systems and their intersection with natural language processing. The KDD Cup 2023 has been a tremendous learning opportunity, showcasing the significance of text features and inspiring my future research direction.

Peiyan Zhang (xu_tai_yu_ in AIcrowd)
I’m a Ph.D. candidate at the Hong Kong University of Science and Technology, with research interests in graph representation learning and trustworthy recommender systems. Participating in the KDD Cup 2023 has been a challenging yet fulfilling experience, encouraging continuous learning and innovation.

Jianxun Lian (master2023 in AIcrowd)
I’m a senior researcher at Microsoft Research Asia, specializing in recommender systems and deep learning techniques. My journey in the KDD Cup 2023 has been a testament to the power of innovation and collaboration in achieving remarkable results.

Defu Lian (Dove in AIcrowd)
I’m a professor at the University of Science and Technology of China, with research interests in spatial data mining, recommender systems, and learning to hash. Participating in the Amazon KDD Cup 2023 has been a rewarding experience, showcasing our research and skills on an international stage.

We Bare Bear

Honghee Lee
I’m Honghee Lee, a Master’s student at Sungkyunkwan University (SKKU) in South Korea, researching knowledge-grounded dialogue systems. Throughout the Amazon KDD Cup 2023, generating next product titles presented intriguing challenges, especially with multilingual issues and cold-start problems. However, exploring various techniques and models helped me gain valuable insights into session-based recommendations in demanding environments.

Youngjae Chang
I’m Youngjae Chang, pursuing a master’s program at Sungkyunkwan University in South Korea. The title prediction task in the Amazon KDD Cup 2023 proved challenging due to the vast number of products and sessions. Despite the complexity, I discovered the power of lightweight methods and simple heuristics in recommendation systems.

Kyuri Choi
I’m Kyuri Choi from South Korea, pursuing a Master’s degree at the Natural Language Processing Lab at Sungkyunkwan University (SKKU). My research primarily centers around few-shot and zero-shot text classification, with a focus on AI ethics and mitigating bias in pretrained language models. Participating in Task 3 of the KDD Cup 2023, the ‘Next Product Title Generation’ challenge, has been a valuable learning experience. I’m sincerely grateful for this opportunity to explore challenging datasets and showcase my passion for AI.

gpt_bot

Senkin13
I’m senkin13, a data scientist with a bachelor’s degree in applied mathematics. Working at DataRobot, I’m deeply committed to solving complex machine learning problems for our customers. Participating in the Amazon KDD Cup 2023 has been a rewarding experience, and I’m thankful for the opportunity to achieve a prize and explore new technologies.

Zhongshan Huang
As a mechanical engineer with a passion for AI technology, participating in various AI challenge platforms has been instrumental in enhancing my skills and knowledge. Being part of the Amazon KDD Cup 2023 was a thrilling experience, competing with some of the best in the field.

zhangweijia
I’m zhangweijia from China, with a Master’s Degree and experience in industrial Recsys. Over the past 5 years, I’ve gained extensive experience in this field and achieved several gold medals in Kaggle competitions. My primary focus is on Recsys, and I’m actively exploring NLP and CV. Participating in the KDD Cup 2023 has been a delightful journey, and I’m grateful for the opportunity to share my interests and experiences.

Conclusion

The Amazon KDD Cup 2023 competition not only united shoppers globally but also pushed the boundaries of AI technology in recommender systems. Through innovative solutions, these brilliant minds demonstrated the power of AI in connecting people and products, regardless of language or cultural barriers. As the datasets are made publicly available, the journey of AI-driven personalization is bound to continue, unlocking new possibilities and reshaping the future of shopping.

Are you excited to be a part of the next AI challenge? Stay tuned to AIcrowd’s upcoming events and follow them on Twitter @AIcrowdHQ for updates. You could be the next champion, revolutionising the world of AI and creating groundbreaking solutions for a global audience.