Abstract: As an important carrier for the implementation of large language models(LLMs), intelligent agent provides a novel path for building a generation new paradigm of academic evaluation. This paper presents the inherent logic of LLM enabled academic paper evaluation, clarifies its indicator design, evaluation mechanism, and application requirements. On this basis, the paper further proposes an overall framework comprising agent tier, model tier, data tier, and interaction tier. Then, we describe the collaborative mechanism among agents and the indicator calculation scheme. Based on these components, we form a method for constructing academic paper evaluation agents that integrate perception, cognition, analysis, and decision-making capabilities. Finally, we systematically implemented the intelligent agent framework and analyzed its effectiveness through application cases. This implementation and analysis demonstrate that the LLM-based agent not only further ensures the availability, credibility, and controllability of the evaluation process, but also represents an exploration of a scalable, adaptive, and intelligent solution for academic paper evaluation.

Key words: Academic evaluation, Paper evaluation, Large language model, Academic agent