More than 99.8% of the declared moderation decisions cite general conditions rather than the law. The categories imposed by Brussels do not lend themselves well to this, according to a study.
Section 24(5) of the Digital Services Act requires online platforms to forward each of their moderation decisions, with a statement of reasons, to the DSA transparency database. Each decision is classified in one of seventeen broad categories. However, more than 99.8% of these declarations invoke a violation of the conditions of use of the service rather than a legal provision: the basis therefore mainly records the application of contractual rules, often classified in tote categories.
31 services screened
Researchers have faced the conditions of use of 31 services designated by the Commission — a total of 447 documents — with the official taxonomy, using a GPT-4.1 few-shot classification. First, the relevance of the categories varies greatly depending on the type of service. Hazardous products account for 17.5% of e-commerce site rules, compared with 0.2% on adult platforms; the negative effects on civic debate weigh 9.9% at Meta but are absent from Booking.com or Wikipedia.
Second, the sub-categories overlap clearly, making it difficult to assign a single category to a decision. The authors believe that the massive use of generic categories stems in part from the design of the scheme, and recommend a differentiated taxonomy by type of service, with more mandatory attributes derived from the actual vocabulary of the platforms.
Read the full article on the original source
Go to the source →