Bloom’s Taxonomy Based Question Analysis for Personalized Learning

In the 1950s, Benjamin Bloom and his associates introduced Bloom’s Taxonomy, a foundational framework for categorizing learning objectives and cognitive skills. While question classification typically operates at a fine level, such as sentences and phrases, and text classification focuses on the document level, past research has explored the intersection of question classification and Bloom’s Taxonomy to assess learners’ cognitive levels in higher education. However, existing feature types from previous studies may excel in datasets with narrowly focused queries, necessitating the development of multiple classifiers for diverse fields or areas. To address this, A new kind of feature called “taxonomy-based” is suggested to improve the accuracy of question classification in datasets from different fields. Utilizing datasets comprising questions from distinct topics, the study evaluates the effectiveness of taxonomy-based features. Support vector machines (SVMs) are chosen as the classifier for their reputed text classification accuracy. The research reveals that taxonomy-based features significantly improve classifier performance when applied to question sets from different domains. This effort is driven by the need to improve question categorization accuracy over a variety of datasets and the realization of Bloom’s Taxonomy’s continuing use in educational settings. Although Bloom’s Taxonomy offers an extensive framework for comprehending cognitive abilities and learning objectives, current question classification techniques frequently do not have the granularity necessary to fully utilize this taxonomy. Furthermore, prior work has mostly concentrated on fine-level categorization, ignoring the advantages of integrating Bloom’s Taxonomy into the classification procedure. The highest accuracy achieved using SVM with TF-IDF vectorizer is 74.17% for BCL’s Data set and 95.23% for BT Dataset. Employing the KNN algorithm with the TF-IDF vectorizer yields 59.17% accuracy for BCL’s Data set, with BT Dataset reaching a maximum accuracy of 89.11%. Naïve Bayes algorithm, coupled with the TF-IDF vectorizer, achieves a peak accuracy of 73.33% for BCL’s Data set and 90.99% for BT Dataset. Lastly, using the Random Forest algorithm with the count vectorizer results in a maximum accuracy of 74.17% for BCL’s Data set and 95.21% for BT Dataset.