Research & Development
With a view to creating new value through cutting-edge technology, we founded ZOZO NEXT, Inc. in 2021. The company is centered around ZOZO Research and MATRIX, our group’s R&D division.
ZOZO Research operates under the mission of “Quantifying Fashion” and conducts research based on the wealth of information we have accumulated through our business. The insights gained from the research are applied to various activities and services including those that utilize AI.
Related Links
ZOZO Research – Our Mission
Service Innovations Driven by Technology and Expertise
The ZOZO Group utilizes various technologies, including AI, to develop new tools and enhance its existing services.
Enabling Smooth Fashion Discovery Through Conversations on ChatGPT
We are among the first in the fashion industry to utilize "Apps in ChatGPT"*, a new feature of ChatGPT.It enables us to seamlessly offer outfit ideas and item information available on the fashion EC site "ZOZOTOWN" directly within ChatGPT. By clarifying user tastes and concerns that are difficult to verbalize and providing information tailored to them, we realize a smoother fashion discovery experience.
*A feature announced by OpenAI in October 2025 and launched in December, which allows users to utilize third-party apps within a conversation in ChatGPT. A connection confirmation prompt is displayed when an app is used for the first time.
ZOZO ー Adds Support for “Apps in ChatGPT” and Begins App Integration
Personalized Outfit Explanations Through Generative AI on WEAR by ZOZO
The fashion coordination app "WEAR by ZOZO" has introduced a new feature called "Mix-and-Match Suggestions." With this new feature, when a user selects a specific item, an AI specialized in fashion suggests outfit ideas tailored to the user's preferences. Furthermore, by inputting their "body shape concerns," users can receive highly personalized suggestions that closely support each individual.
WEAR — An Evolved Fashion-Specific AI — Introduces the New "Mix-and-Match Suggestions" Feature
Development of AI Technology to Decode Ambiguous Fashion Terms
ZOZO NEXT, inc. and a research team from Waseda University have co-developed a new technology that enables AI to quantify ambiguous expressions unique to fashion—such as “casual,” “adult-casual,” and “office-casual”—by analyzing individual item characteristics. This groundbreaking technology is expected to assist users in making more informed decisions across all aspects of their fashion life.
AI assistant That Decodes Ambiguous Fashion Terms
Official Launch of "ZOZOMETRY": A Body Measurement Solution for Businesses
ZOZO, Inc. has launched ZOZOMETRY, a smart measurement solution that generates high-precision 3D data by scanning body with a smartphone. It enables business users to perform body scans more efficiently and reduce related costs through centralized management of scan data. ZOZOMETRY has already been adopted in industries such as custom-made apparel and R&D, and is planned to expand into the medical and fitness sectors.
*A body suit for 3D body scanning
ZOZOMETRY, a Smart Measurement Solution without ZOZO SUIT* for Business
Mixed-Integer Programming for Optimizing Workforce Management
ZOZO Research presents a technique that uses mixed-integer optimization to solve task assignment problems with ZOZOTOWN’s customer support operations as an example. The approach includes minimizing penalties for insufficient processing capacity, applying minimum task duration constraints, and employing a two-stage optimization framework.
A technique for managing scheduling problems by optimizing mixed-integer programming.
Grand Opening of "niaulab by ZOZO ": The Ultimate Personalized-Styling Service
ZOZO, Inc. launched "niaulab by ZOZO," The ultimate personalized styling service that integrates expertise from professional stylists into a custom-built AI system.
Data shows that users who experienced “niaulab by ZOZO” visited ZOZOTOWN approximately 1.5 times more often and spent nearly twice as much as they did before using the service.
*Based on behavioral data from approximately 1,000 users who experienced the service between February 2023 to April 2024, measured over a 30-day period after using the service. New and inactive members at the time of application were excluded from the analysis.
*This service is currently unavailable
"niaulab by ZOZO ": The Ultimate Personalized-Styling Service
Research Highlights
We are mainly engaged in research and development of AI technology.
Innovative Natural Language Processing Research on Static Word Embeddings
A research paper by researchers from ZOZO Research, titled "Static Word Embeddings for Sentence Semantic Representation," has been accepted into the Main Conference of EMNLP 2025 (Empirical Methods in Natural Language Processing), a top international conference in the field of natural language processing.
ZOZO Research Proposes New Method in Natural Language Processing! Paper Accepted at Top Conference "EMNLP 2025"
Releasing Research Results on the Stability and Fairness of Technology for Integrating AI Models
Two research papers co-authored by a research group including researchers from ZOZO Research, titled "DisTaC: Conditioning Task Vectors via Distillation for Robust Model Merging" and "On Fairness of Task Arithmetic: The Role of Task Vectors," have been accepted at "ICLR 2026" (The International Conference on Learning Representations), a top conference in the field of machine learning.
ZOZO Research Has Two Papers Accepted at ICLR 2026, a Top Conference in Machine Learning
Proposing a New Method to Improve Computational Efficiency While Preserving the Expressive Power of Graph Neural Networks.
A research paper co-authored by a research group including researchers from ZOZO Research, titled "Connectivity-Guided Sparsification of 2-FWL GNNs: Preserving Full Expressivity with Improved Efficiency," has been accepted for the Main Track Oral Talks at "AAAI-26," a top conference in the field of artificial intelligence.
ZOZO Research Has a Paper Accepted for Oral Talks at "AAAI-26," a Top Conference in the Field of Artificial Intelligence
A Novel Method for Explaining Black-Box Model Predictions via Dual-Layer Feature Attribution
A research paper co-authored through collaborative research between the Artificial Intelligence and Software Technology Research Center at Chiba Institute of Technology and ZOZO Research, titled "Explaining Black-box Model Predictions via Two-level Nested Feature Attributions with Consistency Property," has been accepted at "IJCAI 2025" (International Joint Conference on Artificial Intelligence), a top conference in the field of artificial intelligence.
Joint Research by Chiba Institute of Technology and ZOZO Research Has a Paper Accepted at IJCAI 2025, a Top Conference in Artificial Intelligence.
New Dataset and Evaluation Metrics for Emotion-Focused Recommendation of Fashion Items
ZOZO Research has developed new datasets and evaluation methods that focus on users’ sentiments—specifically, whether and why they would like or dislike personalized recommendations. These datasets are expected to help reduce mismatches in purchases on e-commerce platforms like ZOZOTOWN.
A paper from ZOZO Research has been accepted to The Web Conference 2025, a leading conference in the field of web and data mining.
Novel Loss Functions Via Applying Mean-Field Theory to Deep Metric Learning
ZOZO Research conducted a study on novel loss functions for efficient learning of similarity between data by applying mean field theory—a technique from statistical physics—to deep metric learning. This research introduced a simpler and more efficient loss function inspired by interaction theory in magnetic materials. Experimental results indicated that image search and recommendation systems performed more accurately when using the proposed loss function.
A paper from ZOZO Research has been accepted to ICLR 2024, a leading conference in the field of AI.
Fashion Suitability: Analyzing the Potential of GPT-4V for Evaluation
ZOZO Research assessed the fashion aesthetic evaluation capabilities of GPT-4V, a multimodal large language model. By comparing its output with a dataset of human evaluations collected from several hundred participants, the study confirmed that GPT-4V’s assessments demonstrated a certain degree of consistency with human judgment.
SIGGRAPH Asia 2024 Technical Communications
Open Innovation
ZOZO Group is committed to research and development to enhance its service quality and deliver a more user-friendly platform. The company also actively contributes to academia and industry through technological innovation.
Open Source Release: ZOZOTOWN’s Fashion Recommendation Data and Algorithm Development Platform
ZOZO Research released "Open Bandit Data & Pipeline" — a large-scale open-source resource consisting of both a dataset and algorithm development platform.
The "Open Bandit Data" is a dataset comprising over 2,600 data units obtained from an actual recommendation algorithm.
The "Pipeline" is an implementation platform that allows users to test new algorithm and evaluate their real-world performance with high reliability, in addition to measuring their accuracy.
ZOZOTOWN’s Fashion Recommendation Data and Algorithm Development Platform Released as Open Source
Open Source Release: Large-Scale Outfit Coordination Dataset and Trend Shift Evaluation Platform
ZOZO Research released "Shift15M, " a large-scale outfit coordination dataset consisting of real data and an accompanying implementation platform as an open-source resource to support research on distribution shift.
It enable users to understand annual trends in outfit coordination and to investigate distribution shifts through reproducible experiments. It is expected to accelerate research in this field by providing a way to capture the ever-changing trends in fashion more accurately.
Large-Scale Outfit Coordination Dataset and Trend Shift Evaluation Platform Released as Open Source
Collaboration with Partners
ZOZO Group works together with a wide range of partners—including companies, universities, and research institutions—to tackle social issues.
Successful Detection of Moderate-to-Severe Scoliosis in Youth
ZOZO, Inc. and a research team from the University of Tokyo successfully detected moderate to severe scoliosis primarily in younger generations which may require treatment using ZOZOSUIT® and a dedicated smartphone application. This approach is expected to contribute to the practical implementation of a new at-home self-screening tool for scoliosis.
Scoliosis Detection Using ZOZOSUIT®
Joint Research on Functional Textiles with the University of Tokyo and Hosoo Co., Ltd.
ZOZO NEXT, Inc., the Yasuaki Kakehi Laboratory at the Graduate School of Interdisciplinary Information Studies of the University of Tokyo, and Hosoo Co., Ltd. have launched a joint research project aimed at developing innovative textiles that dynamically change their appearance and characteristics in response to environmental conditions and computer control. The project combines the advanced techniques of Nishijin weaving with cutting-edge interactive technology.
ZOZO Technologies, Inc., Predecessor of ZOZO NEXT, Begins Joint Research on Functional Textiles with the University of Tokyo and Hosoo Co., Ltd.
Joint Research with Yale University to Expand User Experience and Maximize Customer Lifetime Value for ZOZOTOWN Users
ZOZO, Inc. has initiated a collaborative research project with Yale University (U.S.) to enhance ZOZOTOWN’s service functionality and optimize customer value.
This research seeks to better understand customer incentives and the psychology behind purchasing behavior. By combining Yale’s expertise in quantitative marketing and empirical industrial organization with ZOZOTOWN’s extensive and diverse user data, the project aims to optimize personalized communication strategies for individual users.
ZOZO, Inc. to Launch Joint Research with Yale University Applying Economics to Maximize Customer Lifetime Value for ZOZOTOWN Users
Launching Joint Research with Keio University SFC Research Institute on Textile Design Technology Using Mathematics.
ZOZO NEXT has launched joint research with Keio University SFC Research Institute on textile design technology using mathematics. By combining mathematical theory with our expertise in cutting-edge materials and traditional Japanese textile techniques, we aim to evolve textile design—which has traditionally relied on empirical rules—into a theoretical and reproducible method, and to create new textile technology that combines both functionality and designability.
ZOZO NEXT and Keio University SFC Research Institute Launch Joint Research on Textile Design Technology Using Mathematics: Aiming to Create New Textile Technology-Designing Textile Structures via Mathematical Theory.
Awards
Received the "Annual Conference Award" at the Annual Conference of the Japanese Society for Artificial Intelligence (JSAI2025)
The research titled "Bregman Divergences between Sets and Their Learning via Permutation-Invariant NNs" by researchers from ZOZO Research has received the Annual Conference Award at the "39th Annual Conference of the Japanese Society for Artificial Intelligence (JSAI2025)," one of Japan's largest AI academic events organized by the Japanese Society for Artificial Intelligence (JSAI).
ZOZO Research Receives the "Annual Conference Award" at the Annual Conference of the Japanese Society for Artificial Intelligence (JSAI2025)
First runner-up Presentation Award — the 27th Information-Based Induction Sciences Workshop (ISBS 2024)
The study “Controllable Attractive and Repulsive Forces in Power Set Distributions” conducted by ZOZO Research received the Second Best Presentation Award at the 27th Workshop on Information-Based Induction Sciences (IBIS 2024), Japan’s largest conference in the field of theoretical machine learning.
First runner-up Presentation Award — the 27th Workshop on IBIS 2024
The 42nd Case Study Award — The Operations Research Society of Japan
A presentation by a researcher from the ZOZO Research, titled "Inventory Allocation to Minimize Inter-site Transportation After Expansion of logistics warehouses", has been awarded the 42nd Case Study Award by the Operations Research Society of Japan.※This research is based on a project conducted within the company from March to June 2020. The presentation was made at the Corporate Case Study Exchange Meeting during the Autumn Conference in September 2021.
ZOZO Research Wins the 42nd Case Study Award from the Operations Research Society of Japan for Inventory Allocation Research Minimizing Inter-Facility Transportation