Understanding Knowledge Models: Modeling Assessment of Concept Importance in Concept Maps

Understanding Knowledge Models: Modeling Assessment of Concept Importance in Concept Maps David Leake, Ana Maguitman and Thomas Reichherzer Computer Science Department, Indiana University Lindley Hall 215, Bloomington, IN 47405, USA leake, anmaguit, treichhe @cs.indiana.edu Abstract sites. The CmapTools software has been downloaded by users in approximately 150 countries, and has been used in major educational initiatives, such as the Quorum project [Canas et al., 1995], which involved more than one thousand schools in Latin America. It has also been used for modeling and sharing the knowledge of human experts, for example, for modeling NASA experts’ knowledge of Mars (http://cmex- www.arc.nasa.gov/). CmapTools provides a convenient framework for knowl- edge construction, but users may have dif culty nding rele- vant resources, remembering speci c aspects of a domain to include, or locating relevant concept maps to compare. To alleviate this problem, projects are under way at Indiana Uni- versity and the IHMC to develop intelligent suggesters to support users by retrieving resources such as prior concept maps and multi-media materials [Leake et al., 2003]. Fig- ure 1 shows a screenshot of a Mars knowledge model under construction, with suggestions of propositions, resources, and topics to consider. The suggesters’ effectiveness depends on their ability to retrieve topic-relevant information, which in turn depends on modeling users’ own judgments as they ex- amine concept maps. Thus modeling users’ judgments of the importance of concepts to a map’s topic has practical value for suggester software to support concept mapping and sci- enti c value, for better understanding what in uences human understanding of the knowledge that concept maps convey. The assessment of concept importance may depend on the concepts they include (based on their labels in the concept map), on the concept map topology, or on layout differences between isomorphic maps. Especially for users unfamiliar with a domain, we would expect topology and layout to play an important role in their assessment of the topic of a con- cept map. However, to our knowledge, no previous stud- ies have investigated whether/how the topology and layout of a concept map actually in uence judgments of its topic. To hypothesize candidate topological and layout factors that might in uence decisions of which concepts are most topic- relevant, we considered general structure and layout guide- lines for building good concept maps in the concept map- ping literature, as well as methods for identifying important nodes from the structure of hyperlinked environments. These were used to develop candidate models for the in uence of structural features on identifying the concepts most impor- tant to the topic of a concept map. We then performed exper- iments in which twenty paid subjects judged the relative im- portance of concepts in concept maps selected to investigate particular structural in uences. We used this data to set pa- Concept mapping is widely used in educational and other set- tings to aid knowledge construction, sharing, and comparison; concept maps are also used as a vehicle for assessing under- standing. To aid the concept mapping process, projects at Indi- ana University and the Institute for Human and Machine Cog- nition (IHMC) are developing “intelligent suggesters” to sup- port users as they build concept maps, by presenting them with relevant information from existing knowledge models and the Internet. This depends on identifying important concepts in the concept map under construction. This paper presents and evaluates models of the influence of concept map layout and structure on the selection of concepts expected to be relevant to the topic of concept maps. It presents and assesses a set of potentially-relevant structural factors and evaluates how these factors combine to affect human judgments of concept impor- tance. Twenty subjects were asked to judge the relative impor- tance of concepts in concept maps selected to highlight partic- ular characteristics, and three models were compared to their judgments. Analysis of the results shows that subjects were significantly influenced by concept map topology, but little in- fluenced by other aspects of concept map layout. The results suggest that layout-independent models of concept maps can provide a suitable representation for guiding retrieval of topic- relevant information to support concept map construction, pro- vided that the representation reflects topologically-based influ- ences. The results are applied in the design of the suggesters’ similarity assessment procedures for retrieving relevant con- cept maps. Introduction Concept mapping [Novak and Gowin, 1984] has been widely used to elucidate humans’ knowledge and to facilitate knowl- edge elicitation, construction, and comparison and shar- ing. In concept mapping, users construct a two-dimensional, visually-based representation of concepts and their relation- ships. The concept map representation encodes proposi- tions describing two or more concepts and their relation- ships, in simpli ed natural language sentences. In educa- tional settings, concept mapping exercises have been used to encourage students to actively construct an understand- ing of concepts and relationships within domains of inter- est. To facilitate concept map construction and sharing, the Institute for Human and Machine Cognition (IHMC) has developed CmapTools, publicly-available tools to sup- port generation and modi cation of concept maps in an elec- tronic form (http://cmap.ihmc.us/). CmapTools enable in- terconnecting and annotating maps with material such as other concept maps, images, diagrams, and video clips, providing rich, browsable knowledge models available for navigation and collaboration across geographically-distant