Exploratory Methods for Relation Discovery in Archival Data

TL;DR

This paper introduces a holistic exploratory approach to discover and predict relations in art historical archival data, enhancing cataloguing and biographical enrichment with graph pattern analysis.

Contribution

It presents a novel combination of exploratory data analysis, feature selection, and classification models for relation discovery in archival art data.

Findings

Biographical relations are predicted with higher precision.

Deterministic rules outperform probabilistic methods.

Graph patterns aid in relation discovery.

Abstract

In this article we propose a holistic approach to discover relations in art historical communities and enrich historians' biographies and archival descriptions with graph patterns relevant to art historiographic enquiry. We use exploratory data analysis to detect patterns, we select features, and we use them to evaluate classification models to predict new relations, to be recommended to archivists during the cataloguing phase. Results show that relations based on biographical information can be addressed with higher precision than relations based on research topics or institutional relations. Deterministic and a priori rules present better results than probabilistic methods.