Gyldendal Rettsdata wanted to find out, which kind of sentence was given for which type of narcotic related crimes. The data source for the investigation was Rettsdata’s database, which contains lawsuits and verdicts from more than 40 years.

The project required development of a detection and extraction method for key-features. A key-feature is for example type of narcotic substance, the amount, the illegal act and the sentence. Once the method for key-feature extraction was developed, the same method could also be used to identify and extract other key-features that are not drug related. 

Among other, the method of identifying key-features had to account for the following complexity:
The documentation in the database does not follow a standardized method for quantitative descriptors. E. g. the following units and their abbreviations can be found for quantitative key-features: Grams, Kilogram, g, Kg, “6 – Six – year”, «6 Years», “Six years”, «72 Month», Etc.

In order to deliver relevant key-feature extractions, Nordic Analysis created first training and test data sets based on selected type of documents in the database. Then we designed and trained neural networks for delivering information on amount and type of drug, as well as for illegal act and sentence.
Gyldendal Rettsdata has access to the results via a REST API.