Educational Process Analysis aims to discover latent learning and teaching processes that are hidden within temporal educational data. Typically, the process data has information about what students and teachers do over time. From such data, we can discover different representations of the unobserved learning and teaches processes that might be producing the observed data. In other words, the process data can be thought of as resulting from a set of hidden learning and teaching processes occurring within students and teachers.
To read more about our Process Mining work, check out our publication in the Journal of Educational Data Mining: MODELING NAEP TEST-TAKING BEHAVIOR USING EDUCATIONAL PROCESS ANALYSIS
Processes hidden in the data can be represented by various types of constructs and models. Some examples are directed graphs, process model representations such as Petri Nets, Heuristic Nets, or Fuzzy Nets, graphical models such as Hidden Markov Models or Bayesian Networks, simpler constructs like Markov Chain Transition Matrices, or even trivial representations like discrete event sequences. Techniques such as Association Rule Mining (Garc ́ıa et al., 2010), Sequential Pattern Mining (Zhou et al., 2010), Process Mining (Trcˇka et al., 2010; Bogar ́ın et al., 2018), Graph-Based Analysis (Lynch et al., 2017; Patel et al., 2017), and Curriculum Pacing (Patel et al., 2018) can help us discover different representations of educational processes from the data. For example, the Rule Mining methods can discover which student/teacher interactions follow each other more frequently. Pattern Mining methods can reveal frequent sequences of actions in the data. Process Models and Graph-Based Analysis can give an end-to-end view of the student interaction data as process models or graphs, whereas the Curriculum Pacing method produces a clear visualization of how students follow the curriculum over time.
Educational process data can come in many different shapes and sizes, and we have to use different methods for different types of data. For example, to analyze task-level data with a low amount of variance, we can use graph-based algorithms or process modeling algorithms such as Heuristic Miner (Bogar ́ın et al., 2018). These algorithms become difficult to use when there is a high amount of complexity in the data. This is often the case with click-stream data, where we can use algorithms like Fuzzy Miner (Bogar ́ın et al., 2018) that give more flexibility with ‘zooming in and out’ of the process maps so that we can easily look at both more and less frequent behaviors. If the data have a high amount of variance, meaning that there are too many student learning processes or behaviors tied up with each other, we can use sequence clustering methods to group student data with similar temporal features and analyze them separately (Bogar ́ın et al., 2014; Patel et al., 2017).
Check out another blog post by us that walks through a simple example of doing process mining on educational data: https://www.playpowerlabs.com/post/let-s-do-educational-process-mining
Thank you for reading!