Exploring the Triangulation of Dimensionality Reduction When Interpreting Multimodal Learning Data from Authentic Settings

Abstract

Multimodal Learning Analytics (MMLA) has sparked researcher interest in investigating learning in real-world settings by capturing learning traces from multiple sources of data. Though multimodal data offers a more holistic picture of learning, its inherent complexity makes it difficult to understand and interpret. This paper illustrates the use of dimensionality reduction (DR) to find a simple representation of multimodal learning data collected from co-located collaboration in authentic settings. We employed multiple DR methods and used triangulation to interpret their result which in turn provided a more simplistic representation. Additionally, we also show how unexpected events in authentic settings (e.g., missing data) can affect the analysis results.