Performance Evaluation of Agentic Workflow-Driven Trend-Aware Rule Mining for Dynamic Menu Bundling

Abstract

Digital transformation in the culinary industry currently demands moving beyond writing static lines of code, instead acting as an AI orchestrator adaptive to real-world conditions. This research focuses on addressing significant challenges in traditional data mining methods, such as the Apriori and FP-Growth algorithms, which often lack the flexibility to handle dynamic variables like ambient temperature fluctuations.
Through the innovative orchestration of the Trend-Aware Rule Mining (TARM) algorithm and a LangGraphbased Agentic Workflow, this study transforms raw association rules into strategic business decisions via an iterative reasoning process and self-correction mechanism. Experimental results on a dataset of 52,494 rows demonstrate TARM's computational superiority, with memory usage of only 8.04 MB , significantly more efficient than Apriori's 127.44 MB. Furthermore, the synergy between the Strategy Agent and Evaluator Agent achieved a logic consistency score of 100% , validated by an independent audit with an average score of 96.25%.
These findings confirm that the developed system is in a ready-to-use state to support precise and adaptive decision-making automation in production environments.