Knowledge Graph for Materials Capitalization of Algeria’s Built Heritage: Leveraging Automated Knowledge Extraction and Processing

Abstract

Ancient materials knowledge is a capital resource which ensures the effective preservation and restoration of historical structures. However, such knowledge is often complex to capture due to the nature of interrelated information related to heritage materials. Existing research proposing repositories dedicated for documenting ancient materials as first-class citizens, either proposes manual aproaches, or is commonly centered around relational databases. Contrary to databases, Knowledge Graphs (KGs) are more intuitive, more expressive and more flexible allowing easier extensibilty to other contexts (covering further regions, materials or periods) thanks to their graph structure.

Our study proposes the first KG dedicated to the Algerian ancient materials, and highlights its complete automatic construction pipeline. The generalizability of our approach lies in the modeling of a general KG structure (model) that integrates certain concepts inspired by the materials life-cycle model. Our proposal addresses the dual challenge of ensuring generalizability while managing the domain specificities which involved different challenges related to the existence of long-tail and technical entities, the characteristics of the used language and the content of input sources collected with varying granularity levels. To overcome these challenges, the defined KG required conducting various experiments for testing recent intelligent neural models to illustrate their performances and also their limits for our domain.