Generating conceptual knowledge representations to support students writing scientific explanations

Instructional interventions targeting current student conceptual understandings and built around robust conceptual frameworks can play an important role in science learning. Guided by this principle, my research explores the generation of domain and learner conceptual knowledge representations to support students writing scientific explanations online.;Following a human-centered design approach, my research has been informed by the results of an empirical study with domain and instructional design experts using conceptual knowledge representations to personalize science instruction. My dissertation research contributes novel algorithms for the automatic generation of domain and learner conceptual knowledge maps useful both as embedded computational components to enact pedagogical thinking about a domain and as pedagogical interaction design artifacts. Domain knowledge maps are generated using an innovative multi-document summarization approach to extract pedagogically important concepts from educational digital library resources combined with machine learning techniques to establish links between concepts. Learner knowledge maps are generated using a lexical cohesion analysis of student essays.;The generated domain and learner knowledge maps provide a close approximation of human expert performance in the construction of knowledge maps from educational digital library resources and from student essays as observed in the empirical study. In addition, these knowledge representations appear to effectively support both the computational diagnosis of knowledge deficiencies in student understanding and the design of learner-centered scaffolds to support students writing scientific explanations using multiple online information sources.;The conceptual knowledge representations core to my research enable the design and delivery of instructional technologies capable of targeting current student conceptual understandings through rich pedagogical interactions. Such concept-based instructional technologies may contribute to improving how students learn science and hopefully towards lifting the current condition of science education in the US.