Protecting Consumers from Unfair Contractual Terms Using Natural Language Processing
The client
CLARIN, the Common Language Resources and Technology Infrastructure, is a research consortium of organisations working on natural language processing.
The Polish consumer protection and competition authority wanted to use those capabilities in a system that detects abusive clauses in legal agreements automatically. As part of CLARIN, Netguru worked with a university in Wrocław on a model to prove it can be done.
The challengeWhat the teams actually needed
Almost every online service requires agreeing to terms and conditions. Most people do not read them, and those who do usually lack the legal knowledge to tell whether a clause is valid.
Challenge 01Abusive clauses that look legal
An unlawful clause is often worded almost identically to a lawful one, differing only in small details — so people sign agreements they should not.
Challenge 02Expertise that does not scale
Only legal experts can find illegal clauses, and only by reading documents closely, which cannot be done at the volume consumers actually face.
Challenge 03Proving it is possible at all
The task was to demonstrate that automatic detection could work well enough to protect consumers from corporate abuse.
What we did
A model that finds the unfair clause
Netguru’s NLP team worked alongside researchers at a Wrocław technical university and demonstrated that a system can identify abusive contract clauses in most cases.
Scored properly
The best model reached a macro average F1 of 0.87, which on a zero-to-one scale is a strong result for this kind of task.
Handed over as a benchmark
The dataset went to the client, to act as a benchmark against which future systems can be measured.
A paper together
Netguru and the University are co-authoring a scientific paper so the findings can inform future approaches to recognising unfair contractual terms.
Consumers, protected
The point of the work is protecting people in Poland and elsewhere from agreements they were never in a position to assess.
More like thisOther places we scaled product teams for growth.
What’s next