Cyberspace Administration of China, Ministry of Industry and Information Technology, Ministry of Public Security, and State Administration for Market Regulation jointly promulgated, on December 31, 2021, the Provisions on Administration of Algorithmic Recommendation in the Internet Information Service (互联网信息服务算法推荐管理规定) (the “Provisions”), which became effective as of March 1, 2022.
The Provisions consist of 35 articles that are divided into 6 chapters. Pursuant to Article 2 thereof, the Provisions shall apply to the provision of the Internet information services with the application of algorithmic recommendation technologies (hereinafter referred to as the “algorithmic recommendation service”) within the territory of the People’s Republic of China. The Provisions further define that, “the application of algorithmic recommendation technologies” refers to “the application of any algorithmic technology, including without limitation, generation and synthesis, individualized push, sorting and selection, searching and filtering, and scheduling and decision-making, to provide information to users.”
The Provisions have specified, among others, (1) the algorithmic recommendation service providers’ responsibilities for managing, auditing, and improving the algorithmic recommendation mechanism and services in a manner that would promote positive social values and public welfare, (2) the specific actions such service providers should take to protect the rights and interests of their users; (3) the obligation of an algorithmic recommendation service provider with public opinion attribute or social mobilization capacity to file, via an official electronic record-filing system and within 10 working days following the provision of its services, its name, service type, application field, algorithm type, self-evaluation report of the algorithm, content to be publicized and its other information; and (4) the legal liabilities associated with the violation of the Provisions, which include, in some cases, the suspension of information update, the imposition of a fine of RMB 10,000 – 100,000, administrative penalties, and even criminal liabilities.
Notably, if an algorithmic recommendation service provider commits a violation of any of the following provisions, then criminal liabilities may be imposed if certain thresholds are met. For the avoidance of doubt, the Provisions have imposed other obligations on the algorithmic recommendation service providers, which we will not elaborate here but may elaborate in future blog posts.
- Standards for the Provision of Information Service
Article 7 of the Provisions requires an algorithmic recommendation service provider to fulfill its responsibilities for algorithmic security, to establish and refine the following management rules and technical measures: algorithmic system and mechanism review, science and technology ethics review, user registration, information release review, data security protection and personal information protection, anti-telecom network fraud, security evaluation and monitoring, and emergency plan of security incidents, etc., to formulate and publish the rules relating to algorithmic recommendation service and be equipped with the professionals and technical support commensurate to the scale of algorithmic recommendation service.
Article 8 of the Provisions requires that algorithmic recommendation service providers shall review, evaluate and verify the algorithmic mechanism, model, data and application results, etc. on a regular basis and shall not set up the algorithmic model in violation of any laws, regulations or moral principles such as inducing the users to become addicted or to overspend.
Paragraph 1 of Article 9 of the Provisions requires that algorithmic recommendation service providers shall strengthen its information security management, set up and refine the feature library used for identifying illegal and harmful information, and improve storage standards, rules and procedures. If any information generated or synthesized by the algorithm is found not to be conspicuously marked, the transmission of such information shall be suspended until it is conspicuously marked.
Article 10 of the Provisions states that an algorithmic recommendation service provider shall strengthen the management of user model and user tag and improve the point of interest rules for entry in user model and shall not enter the key words of illegal and harmful information in the point of interest of users or use the illegal and harmful information as user tag and push information content based thereon.
Article 14 of the Provisions states that an algorithmic recommendation service provider shall not use algorithm to register false user accounts, trade accounts illegally, manipulate user accounts or to provide fake likes, comments or forwarding, and shall not use algorithm to block information, provide excessive recommendation, manipulate ranking lists or search results rankings, manipulate the presentation of trending topics, selected contents or other intervening information, to influence online opinions or circumvent supervision and administration.
- Protection of Users’ Rights and Interests
Under Article 16 of the Provisions, an algorithmic recommendation service provider shall, in a prominent manner, inform its users about its provision of algorithmic recommendation service, and notify the public, in an appropriate manner, the basic principles, the purpose and intention, and the main operation mechanism, etc. of its algorithmic recommendation service.
Under Article 17 of the Provisions, an algorithmic recommendation service provider shall provide users with (1) options that are not customized based on the users’ individual characteristics, or (2) the option to conveniently close the algorithmic recommendation service. If a user chooses to close the algorithmic recommendation service, the algorithmic recommendation service provider shall stop providing such service to such user immediately. In addition, an algorithmic recommendation service provider shall provide users with the options to choose or delete user tags customized based on their individual characteristics that are used for providing algorithmic recommendation service. If the application of algorithm by an algorithmic recommendation service provider materially affects the rights and interests of the users, the algorithmic recommendation service provider shall provide explanation and take corresponding responsibility in accordance with the law.
Under Article 22 of the Provisions, an algorithmic recommendation service provider should set up and maintain convenient and effective portals for its users and the public to file claims, complaints and reports of violations, and shall specify the procedures and timeline of its procedures handling such claims, complaints and reports, and shall timely accept and handle such claims, complaints and reports, and provide feedback thereon.
- Record-Filing and Display Requirement
Article 24 of the Provisions requires any algorithmic recommendation service provider with public opinion attribute or social mobilization capacity to file relevant information (e.g., its name, service type, application field, algorithm type, self-evaluation report of the algorithm, content to be publicized, etc.) via an Internet portal in order to complete the record-filing procedures. Such filing shall be conducted within ten working days from the date such service provider starts to provide the relevant services. In addition, if there is any change in the information filed, such service provider shall file such change within ten working days; and if such service provider terminates its services, it shall cancel the record-filing and make appropriate arrangement within twenty working days from the termination of its services.
Article 26 of the Provisions requires an algorithmic recommendation service provider that has completed the record-filing to display its record-filing number and provide a link to the relevant public notification in a prominent place on its websites and application programs through which it provides services to its users.