Leveraging Textual Information for Knowledge Graph-oriented Machine Learning: A Case Study in the Construction Industry

The proven power of Knowledge Graphs (KGs) to effectively represent lexical and semantic information about numerous and heterogeneous entities and their interconnectedness has led to the growing recognition of their potential in engineering disciplines. Meanwhile, a greater focus has been placed on graph embedding techniques to derive dense vector representations of KGs. Such representations could enable the use of conventional machine learning techniques over the content of KGs. However, in the context of building engineering, the quality of the graph embeddings could be problematic, mainly due to the relatively small size of the KGs that are created for individual buildings. This paper aims to investigate the effectiveness of applying KG embedding methods when the elements of the building are described narrowly within the KG. The results of our experiments confirm that proper use of data transformation techniques can significantly improve the quality of the feature representation for downstream tasks.