Scholarlink

<p>Adaptive goal recognition seeks to infer agents’ intentions from observed behavior, yet most existing methods rely on historical traces and therefore fail when newly emerging goals lack prior data. This paper introduces an adaptive framework that integrates process mining with similarity-based goal analysis to enable recognition of unseen goals. New goals are encoded as feature vectors and matched to the most related existing goals using Euclidean distance, Levenshtein distance, and the Jaccard coefficient, allowing surrogate Petri net models to be used until sufficient traces enable dedicated model construction. Experiments across 13 IPC domains and 76 problem categories show that the proposed method significantly mitigates performance degradation: average recall increases by over 22%, yielding an overall recall of 70.95% and balanced accuracy of 69.12%, with a controlled decrease in precision. These results confirm the framework’s effectiveness for early and adaptive recognition of new goals in dynamic and uncertain environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive goal recognition using process mining techniques in dynamic environments and its development to recognize new goals

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Romina Bagheri Mofidi,
  • Fatemeh Davami,
  • Farsad Zamani Boroujeni

摘要

Adaptive goal recognition seeks to infer agents’ intentions from observed behavior, yet most existing methods rely on historical traces and therefore fail when newly emerging goals lack prior data. This paper introduces an adaptive framework that integrates process mining with similarity-based goal analysis to enable recognition of unseen goals. New goals are encoded as feature vectors and matched to the most related existing goals using Euclidean distance, Levenshtein distance, and the Jaccard coefficient, allowing surrogate Petri net models to be used until sufficient traces enable dedicated model construction. Experiments across 13 IPC domains and 76 problem categories show that the proposed method significantly mitigates performance degradation: average recall increases by over 22%, yielding an overall recall of 70.95% and balanced accuracy of 69.12%, with a controlled decrease in precision. These results confirm the framework’s effectiveness for early and adaptive recognition of new goals in dynamic and uncertain environments.