Supporting Collaborative Modeling via Natural Language Processing

Engineering large-scale systems requires the collaboration among experts who use different modeling languages and create multiple models. Due to their independent creation and evolution, these models may exhibit discrepancies in terms of the domain concepts they represent. To help re-align the models without an explicit synchronization, we propose a technique that provides the modelers with suggested concepts that they may be interested in adding to their own models. The approach is modeling-language agnostic since it processes only the text in the models, such as the labels of elements and relationships. In this paper, we focus on determining the similarity of compound nouns, which are frequently used in conceptual models. We propose two algorithms, that make use of word embeddings and domain models, respectively. We report an early validation that assesses the effectiveness of our similarity algorithms against state-of-the-art machine learning algorithms with respect to human judgment.