Text Classification of Software Requirement Documents with TFIDF Models

Abstract

Software requirement documents often contain large volumes of textual statements describing functional behavior, performance limits, security controls, usability expectations, data rules, and interface conditions. Manual classification of these requirements can be slow and inconsistent when documents are written by multiple stakeholders with different terminology and levels of detail. This article presents a TF-IDF-based text classification framework for organizing software requirement documents into predefined categories. The framework prepares a labelled requirement corpus, cleans and tokenizes requirement text, constructs weighted term vectors, trains a supervised classifier, and evaluates category-level performance. The study shows that TF-IDF representation works well when requirement categories contain distinctive vocabulary, especially for security, performance, and interface requirements, but it is less effective when functional, usability, and data-related statements share overlapping terms. The proposed approach supports faster requirement organization, preliminary category assignment, and identification of ambiguous statements that require analyst review.