Learning-Oriented Question Recommendation Using Bloom's Taxonomy and Variable Length Hidden Markov Models

The information overload in the past two decades has enabled question-answering (QA) systems to accumulate large amounts of textual fragments that reflect human knowledge. Therefore, such systems have become not just a source for information retrieval, but also a means towards a unique learning experience. Recently developed recommendation techniques for search engine queries try to leverage the order in which users navigate through them. Although a similar approach might improve the learning experience with QA systems, questions would still be considered as abstract objects, without any content or meaning. In this paper, a new learning-oriented technique is defined that exploits not only the user's history log, but also two important question attributes that reflect its content and purpose: the topic and the learning objective. In order to do this, a domain-specific topic-taxonomy and Bloom's learning framework is employed, whereas for modeling the order in which questions are selected, variable length Markov chains (VLMC) are used. Results show that the learning-oriented recommender can provide more useful, meaningful recommendations for a better learning experience than other predictive models.