Empirical validation of an automated method of knowledge structure creation from single documents

The present study proposes a new method that automatically creates a knowledge structure from a single document. Knowledge structure refers to an organization of knowledge, represented through key concepts that make up the knowledge domain and their proximity relationships. We examine several alternative methods in inferring a knowledge structure from a document, and those methods are compared with the knowledge structures obtained from learners (before and after learning) and domain experts. The knowledge structure created by the method that utilizes paragraph co-occurrences with cosine similarity has been found to be the most successful in generating a knowledge structure from a document and its performance to be comparable to a human expert in terms of similarity and consistency. Findings of the study have the potential to significantly improve the current practice in information retrieval.