Exploring the triangulation of dimensionality reduction when interpreting multimodal learning data from authentic settings

generalizability evaluation

machine learning

multimodal learning analytics

CSCL

collaboration quality

Author

Chejara, P., Prieto, L. P., Ruiz-Calleja, A., Rodríguez-Triana, M. J., & Shankar, S. K.

Doi

https://doi.org/10.1007/978-3-030-29736-7_62

Citation (APA 7)

Chejara, P., Prieto, L. P., Ruiz-Calleja, A., Rodríguez-Triana, M. J., & Shankar, S. K. (2019, Sept). Exploring the triangulation of dimensionality reduction when interpreting multimodal learning data from authentic settings. In the 14th European Conference on Technology Enhanced Learning (EC-TEL) (pp. 664-667). Springer, Cham. https://doi.org/10.1007/978-3-030-29736-7_62

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.