Immaterial damage compensation is a controversial matter in the judicial practice of several law systems. Due to a lack of criteria for its assessment, the judge is free to establish the value based on his/her conviction. Our research motivation is that knowing the estimated amount of immaterial damage compensation at the initial stage of a lawsuit can encourage an agreement between the parties. We thus investigate text regression techniques to predict the compensation value from legal judgments in which consumers had problems with airlines and claim for immaterial damage. We start from a simple pipeline and create others by adding some natural language processing (NLP) and machine learning (ML) techniques, which we call adjustments. The adjustments include N-Grams Extraction, Feature Selection, Overfitting Avoidance, Cross-Validation and Outliers Removal. An special adjustment, Addition of Attributes Extracted by the Legal Expert (AELE), is proposed as a complementary input to the case text. We evaluate the impact of adding these adjustments in the pipeline in terms of prediction quality and execution time. N-Grams Extraction and Addition of AELE have the biggest impact on the prediction quality. In terms of execution time, Feature Selection and Overfitting Avoidance have significant importance. Moreover, we notice the existence of pipelines with subsets of adjustments that achieved better prediction quality than a pipeline with them all. The result is promising since the prediction error of the best pipeline is acceptable in the legal environment. Consequently, the predictions will likely be helpful in a legal environment.
Brazil has a large prison population, which places it as the third country in the world with the most incarceration rate. In addition, the criminal caseload is increasing in Brazilian Judiciary, which is encouraging AI usage to advance in e-Justice. Within this context, the paper presents a case study with a dataset composed of 2,200 judgments from the Supreme Federal Court (STF) about pre-trial detention. These are cases in which a provisional prisoner requests for freedom through habeas corpus. We applied Machine Learning (ML) and Natural Language Processing (NLP) techniques to predict whether STF will release or not the provisional prisoner (text classification), and also to find a reliable association between the judgment outcome and the prisoners' crime and/or the judge responsible for the case (association rules). We obtained satisfactory results in both tasks. Classification results show that, among the models used, Convolutional Neural Network (CNN) is the best, with 95% accuracy and 0.91 F1-Score. Association results indicate that, among the rules generated, there is a high probability of drug law crimes leading to a dismissed habeas corpus (which means the maintenance of pre-trial detention). We concluded that STF has not interfered in first degree decisions about pre-trial detention and that it is necessary to discuss drug criminalization in Brazil. The main contribution of the paper is to provide models that can support judges and pre-trial detainees.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.