Expertsourcing the Identification of Model Elements in Software Model Descriptions

Expertsourcing is a novel approach to model quality assurance: it relies on methods and tooling from crowdsourcing research to split the task of model quality assurance and parallelize its execution across several expert-users. Concretely, given a text-based system description and a corresponding model such as an EER diagram, experts are guided towards inspecting the model based on so-called expected model elements (EMEs). EMEs are entities, attributes and relations that appear in text and are reflected by the corresponding model. Therefore, EMEs play a crucial role in splitting up the model quality assurance task among experts. [Objective & Method] In this paper, we investigate the effectiveness of identifying the EMEs themselves through expertsourcing. To that end, we perform a feasibility study in which we compare EMEs identified through expertsourcing with EMEs provided by a task owner who has a deep knowledge of the entire system specification text. [Conclusions] Results of the data analysis show that the effectiveness of the crowdsourcingstyle EME acquisition is influenced by the complexity of these EMEs: entity EMEs can be harvested with high recall and precision, but the lexical and semantic variations of attribute EMEs hamper their automatic aggregation and reaching consensus (these EMEs are harvested with high precisions but limited recall). Based on these lessons learned we propose a new task design for expertsourcing EMEs.

Authors: Marta Sabou, Dietmar Winkler, Sanja Petrovic

Vorgetragen von: Marta Sabou
Unternehmen: TU Wien

Vortragssprache: Englisch
Level: Fortgeschrittene
Zielgruppe:

Partner 2018