Knowledge Graph Completion Through Ensemble Learning and Incremental KGE Approach Used for Recommendation in Business Process Models

Abstract

Recommendation of a new activity in process models is achieved through the completion of knowledge graph. In this paper we recommend different ensemble learning approaches for completion of knowledge graph thus helping in recommendation in business process. In this process, business process models are first converted to knowledge graphs using translational and embedding approaches and later ensemble learning approaches are introduced to achieve better metrics and reliable recommendations. The ensemble approaches have shown metrics increased for Hits@10 by 32% and Mean Reciprocal Rank (MRR) by 44%. This model is further enhanced with incremental knowledge graph embedding approaches that retain the translation improvements while minimizing the full retraining time. When the process model is extended as a result of addition of new activity, training time is reduced thus speeding up the process 12 times faster with minimal degradation by 4% in metrics.