融合大语言模型与分层语义解析的数学建模论文智能评估系统

An Intelligent Evaluation System for Mathematical Modeling Papers Integrating Large Language Models and Hierarchical Semantic Parsing

摘要: 数学建模论文评估存在人工成本高、维度单一、标准难统一等问题。本文设计并实现了融合大语言模型与分层语义解析的智能评估系统，利用分层语义解析框架来拆解论文结构以及逻辑关系，结合位置感知的数学符号一致性度量模型与符号定义全局链追踪机制，强化对数学建模论文逻辑严谨性的深度评估。进一步引入符号逻辑一致性评分函数与逻辑跳跃判断函数，确保论文中数学符号的一致性与推导过程的逻辑连贯性。再结合大语言模型的上下文理解和推理能力，构建出多维度且可解释的评估体系，实现从文本提取、语义分析到智能评估的全流程自动化。测试结果表明，系统评估结果与专家评分的一致性达89.2%，单篇论文评估耗时 ≤ 45秒，批量处理能力 ≥ 80篇/小时，能够有效支撑数学建模竞赛评审、教学研究等场景的高效评估需求。

Abstract: Evaluating mathematical modeling papers faces challenges such as high manual costs, limited dimensions, and inconsistent standards. This paper designs and implements an intelligent evaluation system integrating large language models with hierarchical semantic parsing. The system employs a hierarchical semantic parsing framework to deconstruct paper structure and logical relationships. Combined with a position-aware mathematical symbol consistency metric and a global symbol definition chain tracking mechanism, it enhances the depth of assessment for logical rigor in mathematical modeling papers. Further integration of a symbolic logic consistency scoring function and a logical leap detection function ensures consistency in mathematical symbols and logical coherence in derivation processes. By leveraging the context understanding and reasoning capabilities of large language models, a multidimensional and interpretable evaluation system is constructed, achieving full-process automation from text extraction and semantic analysis to intelligent assessment. Test results demonstrate 89.2% consistency between system evaluations and expert scores. Single-paper assessment takes ≤ 45 seconds, with batch processing capacity ≥ 80 papers per hour, effectively supporting efficient evaluation needs in scenarios such as mathematical modeling competitions and teaching research.