Services are frequently abusive, as online platforms too often contain in their Terms of Service (ToS) potentially abusive terms for the consumer. Despite the approval of regulations and laws aimed at preventing such situations —like 93/13/EEC Unfair Contract Terms Directive (UCTD) and the Consumer Rights Protection Act in Europe— these problematic terms continue to appear frequently in online ToS.
Limitation of liability clauses restrict the provider's responsibility for damages or losses incurred by the user. These terms may be unfair if they exempt the provider from liability for serious harm, such as physical injuries, intentional damage, or gross negligence.
Termination clauses give providers the right to end the contract or restrict service access. These can be unfair if they allow termination without notice or clear justification, as users may lose access to a service unexpectedly and with no opportunity to remedy any issues.
Unilateral change clauses allow providers to alter the terms of service without the consumer’s consent. This type of provision can be unfair because it binds users to changes they did not agree to, creating an unpredictable and shifting agreement that primarily serves the provider’s interests.
Content removal clauses give providers the right to delete user-generated content. They are potentially unfair if they allow removal without prior notice, clear reasons, or an option for users to retrieve their content.
These clauses state that users accept the terms of service simply by using the platform. This is potentially unfair as it assumes consent without explicit agreement, leaving users bound by terms they may not be fully aware of.
Choice of law clauses specify which jurisdiction’s laws will govern any disputes. They may be unfair if they mandate laws from a region favorable to the provider, often the provider’s own country, rather than the user’s location.
Jurisdiction clauses determine where disputes must be resolved, which can be unfair if they require users to travel to a distant or inconvenient location.
Arbitration clauses require users to resolve disputes outside of court, often through a private arbitrator. These terms can be unfair if arbitration is mandatory and structured to favor the provider, or if it takes place in a distant or unfamiliar location.
The Analyser is a fine-tuned version of BERT, sourced from the LexGLUE project.
This model tackles a multi-label classification problem, where each clause in the ToS can be classified into one or more categories of potentially unfair terms. Instead of generating a single output, the model produces an output per class, each representing a type of potentially unfair clause (e.g., unilateral changes, limitation of liability). Here’s a closer look at how the model works:
These values are also displayed on our interface to indicate how likely it is that a clause fits a specific category, but it’s important to note: this is not an exact percentage or probability and shouldn’t be interpreted as such. Instead, think of it as a relative indicator of how strongly the model believes the clause matches a particular type of unfair term.