Closer Reading of RDF Generated by NLP on Wikipedia Biography: Comparative Analysis

Although Wikidata and DBpedia are closely related to Wikipedia, they often hold a small subset of its semantic information, due to the specific scopes and methodologies chosen for their Linked Data (LD) construction. When we look at biographies, a large amount of RDF statements focus on the person’s core facts, and many semantic narratives are not included. To fill this knowledge gap, this paper seeks a solution with Natural Language Processing (NLP). We aim to assess to what extent out-of-the-box NLP tools can generate new LD from the biographical articles in Wikipedia. Unlike other NLP research, we put more emphasis on the qualitative analysis of NLP outputs (“close reading”) than the statistical performance of NLP algorithms (“distant reading”). We evaluate the overlaps and gaps between Wikipedia, Wikidata and DBpedia, as well as other biographical ontologies. We also analyze the triple patterns from the NLP results in comparison with the RDF entity (instance) and ontologies. Our research revealed that we are able to capture new information about the entity that Wikidata and DBpedia do not hold. At the same time, some noise could not be easily eliminated. Our method presented a bottom-up approach to biographical ontology designing. We also briefly propose a possible solution for future work.