Leveraging Data Augmentation for Process Information Extraction

Business Process Modeling projects often require formal process models as a central component. High costs associated with the creation of such formal process models motivated many different fields of research aimed at automated generation of process models from readily available data. These include process mining on event logs, and generating business process models from natural language texts. Research in the latter field is regularly faced with the problem of limited data availability, hindering both evaluation and development of new techniques, especially learning-based ones. To overcome this data scarcity issue, in this paper we investigate the application of data augmentation for natural language text data. Data augmentation methods are well established in machine learning for creating new, synthetic data without human assistance. We find that many of these methods are applicable to the task of business process information extraction, improving the accuracy of extraction. Our study shows, that data augmentation is an important component in enabling machine learning methods for the task of business process model generation from natural language text, where currently mostly rule-based systems are still state of the art. Simple data augmentation techniques improved the $F_1$ score of mention extraction by 2.9 percentage points, and the $F_1$ of relation extraction by $4.5$. To better understand how data augmentation alters human annotated texts, we analyze the resulting text, visualizing and discussing the properties of augmented textual data. We make all code and experiments results publicly available.

Beyond Rule-based Named Entity Recognition and Relation Extraction for Process Model Generation from Natural Language Text

Automated generation of business process models from natural language text is an emerging methodology for avoiding the manual creation of formal business process models. For this purpose, process entities like actors, activities, objects etc., and relations among them are extracted from textual process descriptions. A high-quality annotated corpus of textual process descriptions (PET) has been published accompanied with a basic process extraction approach. In its current state, however, PET lacks information about whether two mentions refer to the same or different process entities, which corresponds to the crucial decision of whether to create one or two modeling elements in the target model. Consequently, it is ambiguous whether, for instance, two mentions of data processing mean processing of different, or the same data. In this paper, we extend the PET dataset by clustering mentions of process entities and by proposing a new baseline technique for process extraction equipped with an additional entity resolution component. In a second step, we replace the rule-based relation extraction component with a machine learning-based alternative, enabling rapid adaption to different datasets and domains. In addition, we evaluate a deep learning-approach built for solving entity and relation extraction as well as entity resolution in a holistic manner. Finally, our extensive evaluation of the original PET baseline against our own implementation shows that a pure machine learning-based process extraction technique is competitive, while avoiding the massive overhead arising from feature engineering and rule definition needed to adapt to other datasets, different entity and relation types, or new domains.

Evaluating Predictive Business Process Monitoring Approaches on Small Event Logs

Predictive business process monitoring is concerned with the prediction how a running process instance will unfold up to its completion at runtime. Most of the proposed approaches rely on a wide number of different machine learning (ML) techniques. In the last years numerous comparative studies, reviews, and benchmarks of such approaches where published and revealed that they can be successfully applied for different prediction targets. ML techniques require a qualitatively and quantitatively sufficient data set. However, there are many situations in business process management (BPM) where only a quantitatively insufficient data set is available. The problem of insufficient data in the context of BPM is still neglected. Hence, none of the comparative studies or benchmarks investigates the performance of predictive business process monitoring techniques in environments with small data sets. In this paper an evaluation framework for comparing existing approaches with regard to their suitability for small data sets is developed and exemplarily applied to state-of-the-art approaches in predictive business process monitoring.