The subject is complex, but we will use a real example applied here in Softplan, where we use this technology in some of our solutions. An example would be its use in the Gabinete Digital, which is characterized by the automated segregation of elements of legal reasoning (Facts, Orders or other highly representative nuclei) present in judicial pieces. In the case of a 60-page initial petition, for example, an algorithm coupled to the solution suggests to the legal operator (assessor, magistrate or prosecutor) which texts correspond to the requests, the facts or allegations that justify them, and the arguments and norms that mark each request.
The suggestion is recommended based on a knowledge base in which business specialists (internal, from the Lab of Justice, and external, advisors and magistrates) carry out the task of segregating the elements manually. From a predetermined amount of manual extractions, the algorithm, with learning capacity, will suggest – based on the observed patterns of manual extractions -, in an automated way, the extracts of text that correspond to each element of the legal reasoning. The solution user then moves on to the role of trainer of the mathematical model behind the algorithm. By the time he agrees or disagrees and corrects the rating, he is teaching the algorithm about his preference for classification for the process in question.
The better the mathematical model and the more cases are trained, the more assertive the algorithm will be.