engleski

Improving Stability of Detected Topics in Learning Analytics

Due to exponential increase in the number of published research papers, it is becoming more and more difficult to review current research in any discipline. It is especially true for disciplines undergoing fast development like learning analytics and educational data mining. Natural language processing offers opportunity to facilitate extraction of information/knowledge from a large volume of documents. In an attempt to organize papers on application of learning analytics and educational data mining for supporting self- regulated learning, we have used Latent Dirichlet Allocation to identify topics in a set of 107 research papers. Results of Latent Dirichlet Allocation depend on the initial choice of a random number generator seed, and replicated runs do not result in the same topics. Stability of allocation of documents to topics was analyzed using Jaccard similarity coefficient. The proposed method enables identifying stabile clusters of research papers sharing the same topics.

learning analytics ; educational data mining ; self-regulated learning ; text mining ; topic modelling ; Latent Dirichlet Allocation ; Jaccard similarity

nije evidentirano