Business Processes Modeling Recommender Systems: User Expectations and Empirical Evidence

Abstract

Recommender systems are in widespread use in many areas, especially electronic commerce solutions. In this contribution, we apply recommender functionalities to business process modeling and investigate their potential for supporting process modeling. To do so, we have implemented two prototypes, demonstrated them at a major fair and collected user feedback. After analysis of the feedback, we have confronted the findings with the results of the experiment. Our results indicate that fairgoers expect increased modeling speed as the key advantage and completeness of models as the most unlikely advantage. This stands in contrast to an initial experiment revealing that modelers, in fact, increase the completeness of their models when adequate knowledge is presented while time consumption is not necessarily reduced. We explain possible causes of this mismatch and finally hypothesize on two "sweet spots" of process modeling recommender systems.

A process recommendation method using bag-of-fragments

“…Deng et al (2016) presented a process recommendation system that can recommend proper nodes based on patterns mined from existing process repository, where three different recommendation strategies were designed. Fellmann et al (2018) proposed a recommendation-based business process modelling approach, where the goal is to get the evidence of how well process modelling recommender systems might ease daily work of modelers.…”

Augmenting Modelers with Semantic Autocompletion of Processes

“…Commercial companies and the public sector alike have adopted process models as a means for visualizing and executing their business logic. As the volume of existing process models in an organization increases, it becomes evident that reuse and automation are paramount to enable faster design of high-quality models [11,12].…”

“…Several past attempts to solve the autocompletion problem focused on syntactic information, such as the structure of process model graphs, and did not take into account the semantic meaning of the processes and their fragments [5,12,15,23,35]. Other researchers investigated semantic-based approaches [19,20,31,33], however, semantic similarity between sub-processes determined with modern deep learning techniques has not been fully explored yet.…”