Using Automatically Generated Concept Maps for Document Understanding: a Human Subjects Experiment

Concept maps present information in a concise and easily-understood form. Consequently, concept maps of documents are a useful vehicle for summarizing the documents’ contents. Concept-map based summaries can in turn be used as the basis for browsable indices to help guide navigation through documents to find material of interest. However, using concept maps in this role depends on the ability to generate a concept map for each document. Especially for large document sets, constructing the needed concept maps by hand may be prohibitively expensive. This paper addresses this problem with an algorithm to automatically generate concept map fragments to aid in document understanding. The algorithm was evaluated with a human subjects study assessing the value of its results to facilitate locating and understanding portions of a document of interest. The study compared subjects’ speed and accuracy in answering questions about material contained in a document when using only the document, a manually constructed concept map, or an automatically-generated concept map fragment. The study showed that providing automatically generated concept maps improved user speed while retaining accuracy, for documents whose size enabled capturing most key concepts in a single concept map..