Assessing book-length reading comprehension through knowledge structures.

Abstract

Book-length reading requires readers to continuously maintain, update and integrate information across extended texts. However, traditional comprehension assessments of book-length reading provide limited insight into readers' ability to organize textual information globally. The present study examined whether knowledge structures can be used to assess students' comprehension of book-length texts and to evaluate the cross-genre applicability and diagnostic value of this approach. Knowledge-structure representations were collected from 209 students after they read a narrative novel or an expository science book. Bayesian regression models were used to predict reading comprehension. Additional comparisons were conducted to examine differences in knowledge structures between students with different levels of comprehension. Knowledge structures measured by semantic networks predicted book-length text comprehension in both genres. At the same time, the knowledge structures constructed by students after narrative reading tended to reflect a sequence-based structure organized by temporal and causal relations, whereas those constructed after expository reading tended to reflect a hierarchy-based structure organized by logical relations among concepts. In addition, high-performing readers constructed networks with higher transitivity and clustering coefficient, and greater similarity to the expert reference network, indicating stronger structural organization. Low-performing readers' networks showed weaker or more fragmented organization and lower similarity to the expert reference network. This study extends the assessment based on knowledge structure to book-length reading and shows that semantic networks can capture the structural quality of students' understanding after extended reading. This approach offers a promising framework for diagnostic assessment and targeted reading instruction.