Legal Judicial Outcome Classification (JOC) Dataset

This dataset was developed to support research on Judicial Outcome Classification (JOC) in legal Natural Language Processing (NLP). It consists of manually annotated legal text segments extracted from the Supreme Court Judgments (India, 1950–2024) dataset available on Kaggle. The dataset was created because no publicly available benchmark provides sentence-level judicial outcome annotations suitable for transformer-based legal document analysis.The dataset contains 7,500 legal text segments, equally distributed across three judicial outcome categories: Positive (2,500 samples), Negative (2,500 samples), and Neutral (2,500 samples). Each text segment represents a meaningful portion of a legal judgment describing judicial reasoning, legal findings, or case outcomes.Annotation GuidelinesThe dataset was manually annotated by legal domain experts according to predefined judicial outcome annotation guidelines.Positive: Judgment text indicating an outcome or legal reasoning favorable to the claimant or petitioner.Negative: Judgment text indicating an outcome unfavorable to the claimant or petitioner or favorable to the respondent.Neutral: Judgment text containing procedural information, factual descriptions, or balanced legal reasoning without a clearly favorable or unfavorable outcome.The annotation quality was assessed using Cohen's Kappa (κ = 0.85), indicating strong inter-annotator agreement.Dataset CharacteristicsDomain: Legal NLPLanguage: EnglishSource: Supreme Court Judgments (India, 1950–2024)Total Samples: 7,500Classes: Positive, Negative, NeutralClass Distribution: Balanced (2,500 samples per class)Annotation Type: Manual expert annotationInter-Annotator Agreement: Cohen's κ = 0.85