Press Releases

November 21, 2025

DOCOMO Wins Gold Medal in Kaggle, the World's Largest AI Data Science Competition

  • Print

TOKYO, JAPAN, November 21, 2025 --- NTT DOCOMO, INC. announced today that a team including DOCOMO employees, Meisaku Suzuki and Kenichiro Miyaki, won Gold Medal in the international competition “Jigsaw – Agile Community Rules Classification”,*1 hosted on Kaggle,*2 one of the world's largest AI data science competition platforms. The competition was held from July 2025, and the team ranked 11th out of 2,445 participating teams.

Mr. Suzuki has also been awarded the title of “Kaggle Master”,*3 a distinction held by only about the top 1% of the approximately 200,000 active Kaggle competitors*4 worldwide.

From left: Meisaku Suzuki, Kenichiro Miyaki

Kaggle is an online community and platform for data scientists and machine learning engineers, with over 27 million users.*3 It hosts competitions where participants apply cutting-edge algorithms to real-world problems and compete based on the accuracy of their solutions. This time, the competition focused on text classification under community rules. Participants were tasked with determining whether comments posted on the online forum Reddit*5 violated each community's guidelines, such as bans on advertising.
The DOCOMO team addressed this challenge by leveraging multiple AI models to interpret the contextual nuances of both the rules and comments, achieving highly accurate predictions.

DOCOMO is actively developing AI-driven technologies, including large language models and the big-data–based customer-insight engine “docomo Sense®”.*6 Going forward, DOCOMO will continue to encourage employee participation in competitions as part of its commitment to nurturing data-science talent, while leveraging these skills and insights to deliver new value to customers.

  1. New windowhttps://www.kaggle.com/competitions/jigsaw-agile-community-rules
  2. New windowhttps://www.kaggle.com/
  3. Kaggle ranks participants based on the number of medals earned; the second-highest rank, Kaggle Master, requires at least one gold medal and two silver medals.
  4. As of November 21, 2025.
  5. New window https://www.reddit.com/
  6. New windowhttps://ssw.web.docomo.ne.jp/marketing/strengths/sense/ (only in Japanese)

docomo Sense® is a registered trademark of NTT DOCOMO, INC.

Appendix

Overview of the Competition

  • 1. Theme
    The theme of the competition was to classify whether comments posted on the online forum Reddit violated community rules. Participating teams devised methods suited to the provided dataset and competed based on the accuracy of their text classification.

    Provided Dataset (Key Items):

    • Community rules: Text of rules for 6 communities
      Example: No advertising, no legal advice, etc.
    • Positive samples: Texts that violate the community rules
    • Negative samples: Texts that did not violate the community rules
    • Violation labels: Binary values (0 or 1) indicating whether a text violates a rule

    Task:
    As shown in Figure 1, the competition required teams to determine whether each Reddit comment violated any of the six community-specific rules.

    Moderator: Determine whether the post violates community rules. User Post Comment. Online forum; Community rule: No advertising. Community rule: No legal advice. Moderator: Determine whether the post violates community rules. User Post Comment. Online forum; Community rule: No advertising. Community rule: No legal advice.
    Figure 1. Problem setting of the competition
  • 2. Proposed Solution
    The DOCOMO team, including Meisaku Suzuki and Kenichiro Miyaki, developed an approach in which AI models were trained online to predict whether a given sample would violate community rules if it were posted on Reddit. Thirteen AI models with different areas of expertise were used, and their predictions were combined to improve accuracy. The team also applied techniques such as pseudo-labeling, reusing predicted labels to further train the models, and, for some models, creating separate AI models for each community rule. These strategies enabled highly accurate predictions (see Figure 2).

    AI models used:

    • Large Language Model (LLM): AI specialized in predicting the next word in text
    • Embedding models*1: AI that excels at capturing semantic similarity between texts
    • Classification models*2: AI specialized in classifying text
    Comments (Training & test data). Online learning: Large language model, Embedding model, Classification model. Ensemble of AI predictions. Final prediction.
    Figure 2. Gold medal–winning solution
  1. Models that convert text or image data into numerical vectors
  2. Machine learning models that classify input data into predefined categories
Go to top of the page