With rapid advancements in large language models for natural language processing, their role in semantic-aware process mining is growing. We study semantics-aware process mining, where decisions must reflect both event logs and textual rules. We propose an online, adaptive multi-agent framework that operates over a single knowledge base shared across three tasks—semantic next-activity prediction (S_NAP), trace-level semantic anomaly detection (T_SAD), and activity-level semantic anomaly detection (A_SAD). The approach has three key elements: (i) cross-task corroboration at retrieval time, formed by pooling in-domain and out-of-domain candidates to strengthen coverage; (ii) feedback-to-index calibration that converts user correctness/usefulness into propensity-debiased, smoothed priors that immediately bias recall and first-stage ordering for the next query; and (iii) stability controls—consistency-aware scoring, confidence gating with failure-driven query rewriting, task-level trust regions, and a sequential rule to select the relevance–quality interpolation. We instantiate the framework with Mistral-7B-Instruct, Llama-3-8B, GPT-3.5, and GPT-4o and evaluate it using macro-F1. Compared to in-context learning, our framework improves S_NAP, T_SAD, and A_SAD by 44.0%, 15.6%, and 7.1%, respectively, and attains the best overall profile against retrieval-only and correction-centric baselines. Ablations show that removing index priors causes the steepest degradation, cross-task corroboration yields consistent gains—most visibly on S_NAP—and confidence gating preserves robustness to difficult inputs. The result is immediate serve-time adaptivity without heavy fine-tuning, making semantic process analysis practical under drift.

Recent advancements in generative AI have improved process mining, making workflow analysis more accessible and scalable. However, large language models lack structured reasoning and fail to capture sequential dependencies in workflows, limiting their effectiveness. While retrieval-augmented generation (RAG) improves contextual knowledge integration, it does not enforce process execution constraints, leading to inconsistencies in workflow modeling. To address this, we propose GRAG4PM, a graph retrieval-augmented generation framework designed for process mining. GRAG4PM introduces three key innovations: a hierarchical graph architecture that captures multi-level process semantics, an adaptive pruning mechanism that refines workflow representations while preserving critical information, and a process-specific dual-indexing scheme that adapts existing hybrid retrieval to the constraints of workflow graphs. Experimental results demonstrate that GRAG4PM significantly improves workflow validation, anomaly detection, and predictive analysis, particularly in handling incomplete information. Across the three evaluated process-aware tasks, workflow-state validation, violation checking, and next-activity prediction, GRAG4PM achieves an average Macro-F1 score of 0.7259, outperforming Graph RAG (0.4394) and standard RAG (0.4036). Under missing-attribute settings, GRAG4PM also maintains the highest average Macro-F1 across masking conditions, confirming its robustness in incomplete data scenarios. By balancing structured workflow constraints with flexible AI-driven retrieval, GRAG4PM enables dynamic, interpretable, and adaptive process mining solutions and explores promising directions for future research on improving workflow solutions for AI.