Enterprise Modelling Assistance: Edge Prediction Improvement Using Textual Information

Abstract

Today, enterprise modelling is still a highly manual task. There are exist some assistance techniques but they are mostly limited to pattern libraries and pre-defined rules, which limits their functionality and makes them non-flexible. In our previous work we proved the applicability of machine learning techniques to the enterprise modeler support. However, one of the main problems in this area today is the absence of model repositories that could be used for training what causes the necessity to train machine learning models on small data. In this paper we study which textual information from the model and how can be used to increase the efficiency of the edge prediction task, which is one of the key tasks in graph-structured problems like enterprise modelling. The comparative analysis shows that application of FastText method provides a better result for node names embedding, and consideration of node names and descriptions significantly increases the edge prediction quality. The built model is successfully validated on a test case scenario simulating the enterprise model building process.

Keywords

• enterprise modelling
• edge prediction
• text embedding
• graph neural networks