A Review of Modern Requirement Prioritization: From NLP Pipelines to Agentic AI

Requirement Prioritization (RP) and Issue Prioritization (IP) are critical for effective software engineering, dictating optimal resource allocation and deployment sequencing. While conventional frameworks such as the Analytic Hierarchy Process (AHP) and MOSCOW offer structured approaches, they lack the scalability required for modern Agile and DevOps environments. To evaluate the technological trajectory toward automated backlog management, this study conducts a PRISMA-compliant Systematic Literature Review (SLR) analyzing the paradigm shift from foundational Natural Language Processing (NLP) pipelines to Large Language Models (LLMs) and autonomous agentic workflows. Synthesizing data from exactly 78 peer-reviewed empirical studies published between 2015 and 2025, the review quantifies the evolution of prioritization efficacy. The analysis reveals that while static NLP models achieve robust baseline metrics demonstrating average top-3 accuracy scores of 81% and Mean Squared Errors of 2.2 on massive datasets exceeding 29,000 repository issues, they inherently lack deep contextual inference capabilities. Conversely, recent LLM-augmented and deep learning frameworks demonstrate prioritization accuracies ranging from 73% to 90% while reducing processing times to under 25 seconds. Optimized metaheuristic approaches additionally report up to a 30% performance increase over traditional AHP. However, despite these statistical improvements, enterprise LLM deployments remain constrained by hallucination risks and an inability to adapt to real-time temporal dependencies without human intervention. Addressing this critical gap, the review examines the orchestration of LangChain-based agentic architectures capable of dynamic, self-correcting multi-agent decision logic. The study concludes by presenting an agentic workflow framework validated against historical issue tracking datasets and proposes a consolidated research agenda focusing on algorithmic governance, multi-stakeholder fairness, and real-time dependency graph modeling in next-generation automated requirement engineering.