[Class Method Co., Ltd.] Achieved automatic spam detection using generated AI, reducing inappropriate posts by approximately 90% compared to the previous month
*Class Method Co., Ltd.*
Press release: September 19, 2024
**
Achieved automatic spam detection using generated AI, reducing inappropriate posts by approximately 90% compared to the previous month
*Streamline the development and operation of the technology blog “Zenn” with 120,000 members with LLM*
Class Method Co., Ltd. (Headquarters: Minato-ku, Tokyo, Representative Director: Yokota
Satoshi (hereinafter referred to as “Class Method”) is pleased to announce that we have achieved measures against spam posting on the information sharing platform “Zenn” using generative AI.
[Background]
Zenn, an information sharing platform operated by Class Method, has over 120,000 members writing technical blog articles and receives over 13 million PVs (number of page views) per month.
Since the number of spam posts (*) has increased since June 2024, we have considered anti-spam measures to protect the user experience of sharing technical information.
Currently, Zenn receives about 1.11 million characters worth of posts per day, such as blog articles and comments, and it is not realistic to check all of them manually. Assuming that the number of members and posts will continue to increase, automating content checking was essential.
* Directions to sites unrelated to blog content or IT technology, meaningless character strings, etc.
[About spam posting measures using generated AI]
In the class method, we have implemented a mechanism that
automatically detects content that is suspected to be spam, such as articles and comments posted on Zenn, and notifies the administrator. Operation began in early July 2024, and approximately 1,500 cases were automatically detected in August.
We used LLM (Large-Scale Language Model) to determine spam, and were able to develop and implement a detection mechanism with a certain degree of accuracy in a short period of time.
Content that is suspected to be spam is subject to a final human check, and appropriate measures are taken, such as making the article private or freezing the account. Although it is not possible to make a simple comparison as the number of spam posts varies greatly from month to month, after thorough countermeasures from July to August, the number of suspected spam posts in September was reduced by approximately 90% compared to the same period of the previous month (*). The number of complaints from users has also decreased
significantly.
* Estimated as of September 11, 2024
Flow of content patrol using generated AI, automatic spam
detection/reporting, and content confirmation by administrator
Regarding the implementation details, the developer has written and published an article on Zenn.
[Reference] Developer blog: “Spam postings to Zenn increased rapidly, so LLM did something about it”
https://zenn.dev/team_zenn/articles/zenn-check-spam-by-llm
[About “Zenn”]
Zenn is a technical blogging platform operated by Class Method that aims to create a new information sharing community for engineers. In addition to personal use, it is also used by companies, organizations, and teams, and as of September 2024, the number of members has exceeded 120,000, the number of corporate and corporate uses has exceeded 900, and the monthly PV has exceeded 13 million. I am. Zenn members can not only post their findings as articles, but also compile them into a book and publish them for free or for a fee, or post short notes before compiling them into an article. Readers can support authors by gifting badges for helpful articles or by purchasing books.
“Publication” is a function that allows you to create media on Zenn that can be managed by multiple people. By creating a publication for a company, technical community, etc., multiple people can post articles on a specific theme in one place. In order to encourage the creativity of authors, we have a unique system in which articles submitted to a company’s publication remain the author’s own. We have also launched a usage support menu (https://zenn.dev/biz-lp) for companies and corporations, contributing to the creation of a diverse technical community.
[Reference] Zenn, an information sharing community for engineers, has over 100,000 members and 500 organizational users.
https://classmethod.jp/news/20231212-zenn-members-100k/
Class Method will continue to work hard to operate Zenn so that engineers can share information and interact comfortably.
[About class methods]
Class Method Co., Ltd. is Amazon Web
We provide technical support for companies in the fields of services, data analysis, mobile, IoT, AI/machine learning, etc. AWS Support has been recognized as the top partner since 2015, and the “AWS Service Partner” award is given to the best partners in Japan.
of the Year” three times. In 2022, SI Partners of the
The company won the 2016 award, and was a finalist again in 2013, making it one of the world’s top AWS partners in both name and reality.
To date, we have provided technical support to approximately 4,000 companies and over 25,000 accounts in the AWS environment. We also focus on disseminating technical information by our employees, and have published over 40,000 articles on our owned media “DevelopersIO.” Based on our corporate philosophy of “continuing to contribute to the creative activities of all people,” we will continue to propose the most suitable technologies.
Head office location: 26th floor, Hibiya Fort Tower, 1-1-1
Nishi-Shinbashi, Minato-ku, Tokyo
Representative: Representative Director and President Satoshi Yokota Official site: * https://classmethod.jp/* Technical blog “DevelopersIO”: * https://dev.classmethod.jp/*
Facebook page: * https://www.facebook.com/classmethod/*
Official YouTube channel: https://www.youtube.com/c/classmethod-yt/
*The company names and product names listed are registered trademarks or trademarks of each company.