A model-based on embeddings and contextual analysis for relationship recognition in noisy historical writings

Extracting relationships from historical writings to create networks of people may reveal hidden connections. People are frequently referenced directly or indirectly in historical writings or inscriptions (seals, tombstones, stelae, columns, or other historical objects). By mapping out who interacted with whom using several writings, we can reveal relationships not apparent in a single source. This paper presents a method for entity relationship detection in historical writings based on semantic vector representations, contextual analysis, and association rules. Our proposal has the potential to detect relations between entities mentioned in distant unrelated inscriptions, as evidenced by some experiments conducted on historical writings. These results can also be extended to detect relationships between persons, places, roles, or events in historical contexts.