Process Mining-Based Goal Recognition System Evaluation Dataset

These datasets are used for evaluating the process mining-based goal recognition system proposed in the paper "Fast and Accurate Data-Driven Goal Recognition Using Process Mining Techniques." The datasets include a running example, an evaluation dataset for synthetic domains, and real-world business logs.

running_example.tar.bz contains the traces shown in figure 2 of the paper for learning six skill models toward six goal candidates and the three walks shown in figure 1.a.

synthetic_domains.tar.bz2 is the dataset for evaluating GR system in synthetic domains (IPC domains). There are two types of traces used for learning skill models, generated by the top-k planner and generated by the diverse planner. Please extract the archived domains located in topk/ and diverse/. In each domain, the sub-folder problems/ contains the dataset for learning skill models, and the sub-folder test/ contains the traces (plans) for testing the GR performance. There are five levels of observations, 10%, 30%, 50%, 70%, and 100%. For each level of observation, there are multiple problem instances, the instance ID starts from 0. A problem instance contains the synthetic domain model (PDDL files), training traces (in train/), and an observation for testing (obs.dat). The top-k and diverse planners for generating traces can be accessed here. The original PDDL models of the problem instances for the 15 IPC domains mentioned in the paper are available here.

business_logs.tar.bz is the dataset for evaluating GR system in real-world domains. There are two types of problem instances: one with only two goal candidates (yes or no), referred to as "binary," and the other containing multiple goal candidates, termed "multiple." Please extract the archived files located in the directories binary/ and multiple/. The traces for learning the skill models can be found in XES files, and the traces (plans) for testing can be found in the directory goal*/.