Mining Sequential Patterns in Student Collaborative Dialogue While Learning From Research Quest

• In Event: 41.025 - Learning, Engagement, and Assessment in Technology-Rich Learning Environments

Abstract

Objective. The Cleveland-Lloyd Dinosaur Quarry (CLDQ) Research Quest is a digitally-supported, inquiry learning environment designed to support students' critical thinking and analysis of scientific evidence. The goal of this Research Quest is to engage students with authentic scientific objects as they work collaboratively to resolve a scientific question or controversy. The results reported here are from the first investigation of the CLDQ Research Quest, where students worked collaboratively to gather evidence and to develop an argument about the type and species of dinosaur represented by a small set of fossils from the NHMU. We demonstrate feature selection and extraction methods for sequential pattern mining that allow researchers to discover useful information from the discourse moves that characterize the group discussion.
Methods. Four 6th grade classrooms in urban, Title 1 schools in the western U.S. participated in two classrooms sessions of 90 minutes. All four classrooms were taught by different teachers; a total of 104 students participated in the research (average class size = 26). Each classroom was split into four collaborative groups; students worked in groups of 4-7, depending upon class size. During a total of three 90-minute sessions, science educators and project personnel led the classes and facilitated a student-led inquiry process. The data presented in this paper are drawn from the first 30 minutes of the initial classroom session, where student groups began their investigations by making observations about the 3D prints of the three “mystery fossils” to be evaluated. Audio recordings were professionally transcribed, segmented, and coded for the presence of major critical thinking processes in science (see Butcher, Runburg, Altizer, in press), in addition to codes related to instruction (teacher input) and the task.
Results. We identified patterns of discourse moves associated to students’ observations based on several interestingness metrics, as shown in Table 1. The findings suggest that student observations of fossils are most often elaborate rather than superficial. Furthermore, elaborate observations most often occur in series (LR(Observing +t → Observing +t+1) = 0.12, t(13) = 6.37, p < .001, d = 1.70). The utterances of both superficial and elaborate observations were segmented using text pre-processing algorithms, categorized according to a K-Means divisive clustering method, and classified using Bayesian statistics to ease interpretation. Elaborate observations showed a greater degree of lexical diversity than superficial observations as measured by the amount of unique nouns (felaborate = 130 & fsuperficial = 36, respectively). Figure 1 shows that the divisive clustering algorithm required a greater amount of partitions in order to achieve substantial agreement (i.e., Cohen’s k > 0.81) in classifying the clusters of student observations, resulting in fine-grained sub-categories of elaborate observations (Min. Splitelaborate = 7 & Min. Splitsuperficial = 1, respectively).
Significance. The inferences drawn from the examination of sequential patterns in dialogue moves are often complicated due to many features that are irrelevant or ambiguous. The feature selection and extraction techniques for sequential pattern mining utilized in this study stands to improve the understanding of dialogue processes and enables the detection of important sequential patterns.

Authors

• Eric G. Poitras, University of Utah

• Kirsten R. Butcher, University of Utah

• Michelle Hudson, University of Utah