On Friday, the Cyber Administration of China (CAC) issued draft guidelines on regulating the recommendation algorithms used by internet service providers, stating that they should not be used to withhold information, over-recommend or manipulate relevant record charts.
In addition, relevant firms must provide users with convenient options for turning off algorithmic recommendation functions and “immediately” implement any requests to opt out, the CAC said, adding that the draft legislation is open for public feedback until September 26.
The draft rules also list various behaviors prohibited by other algorithmic recommendation service providers. For example, algorithms should not be used to create fake user accounts, engage in illegal trade or manipulate accounts. False praise, commenting, sharing and web navigation caused by algorithmic recommendation shall also be prohibited. Moreover, algorithms shall not be used to manipulate the sorting of search results or to control trending topics.
Recently, a string of domestic social events, such as Kris Wu being arrested for rape and Zhang Zhehan being blocked by media platforms for misconduct, have aroused heated discussion on China’s Twitter-like Weibo. Some netizens have questioned the honesty of spending money to withdraw negative topics from the public arena. On August 23rd, in response to the above rumors, the administrator of Weibo announced management rules for trending topics on Weibo, and stressed that there was no commercial trading involved.
Earlier today, the CAC also announced that it would require platforms to remove various popularity lists from entertainment platforms. Under this new policy, it is strictly forbidden to add or disguise such lists and related products or functions.
Algorithmic recommendation service providers should establish management systems to monitor user registration and information release, and ensure data security and personal information.
Internet companies globally use algorithms to predict user preferences and make content recommendations. For example, a short video platform will label each user according to the background algorithm combined with big data. If the user prefers videos in one specific field, it will give priority to pushing similar videos to the user, so as to enhance the user’s affinity to the product.
The draft legislation released Friday proposes that the algorithmic recommendation service providers should not record illegal or bad information as keywords or use them as user tags. Similarly, discriminatory or biased tags should be banned. Users, meanwhile, should have the right to select, modify or delete all of these tags, whether positive or negative.