ResolveLLM: A Large Language Model Framework for Intelligent Conflict Detection and Resolution in Collaborative Software Development

This paper presents ResolveLLM, an advanced framework for intelligent conflict detection and resolution in collaborative software development. Built on large language models (LLMs), ResolveLLM addresses a significant challenge in modern software engineering: resolving both syntactic and semantic merge conflicts in version control systems. Traditional conflict resolution tools typically struggle with understanding the intent behind changes, leaving developers to manually resolve conflicts, which is time-consuming and error-prone. ResolveLLM automates this process by utilizing transformer-based models trained on extensive code corpora, enabling the system to detect and resolve conflicts with high accuracy. Our experiments show that ResolveLLM outperforms existing models, achieving a conflict detection accuracy of 95.1% and an $\mathbf{F 1}$-score of 95.2%. Additionally, the system demonstrates superior performance in conflict resolution tasks, including a 54.3% exact match rate and a 92.5% compilation success rate.