HUMAN RIGHTS CASE ANALYSIS USING AI AND TRANSFORMER MODELS

Abstract

The growing number of human rights cases requires smart systems that will be able to analyze legal papers in the most effective and correct way. Manual review is time-consuming, laborious and prone to human error and thus difficult to identify trends or commonality among large amounts of data. The current project is based on the AI methodologies, namely, Transformer-based Legal-BERT and Doc2Vec with Support Vector Machine (SVM), which are used to automate the processing of human rights case documents. The methods allow semantic interpretation, feature extraction, and classification of the legal texts, helping the stakeholders find violations and find the relevant precedents faster with high-quality case document embeddings generated by Legal-BERT that is a variant of BERT trained on legal corpora. The model assists in classifying of cases, violation detection and semantic similarity analysis tasks where it is possible to find similarities between cases and common patterns in the violation of human rights. Parallel to that Doc2Vec uses full documents to calculate representations in the form of vectors which are later classified with the help of SVM to make predictions about the possible case results or its categories. This hybrid method will be beneficial in improving overall performance by delivering deep contextual embeddings and strong traditional machine learning to analyze the cases of human rights and make it scalable and constantly learning with the addition of new cases. Combining Legal-BERT and Doc2Vec + SVM, the project will create a multifunctional AI-based system of processing legal documents, identifying patterns, and decision support and increase the efficiency of legal professionals, NGOs, and policy-makers in defending human rights.