A Study on Text Mining and Educational Data Analytics for Assessing Academic Language Competence in Higher Education

Abstract

The ability to use academic language is an important aspect influencing the success of students in higher education, although its assessment still remains laborious due to reliance on manual approaches which are not easy to implement at a large scale. This paper presents a theoretical framework that makes use of text mining and educational data analytics in order to develop an automated approach to assessing the proficiency in academic language of university learners. In the framework under consideration, various textual artifacts created by students (essays, discussion board messages, short answer questions) are used along with behaviorally related data acquired from learning management systems. The proposed approach includes several stages such as data collection, text pre-processing, linguistic feature extraction, analytics modeling, evaluation, and feedback generation. Lexical, syntactic and semantic features are extracted using traditional natural language processing tools and contextual representation approaches, whereas educational data mining techniques are used for learner engagement analysis. The system involves the integration of supervised machine learning algorithms such as support vector machines and random forests alongside deep learning and transformers including Bi-LSTM and BERT/RoBERTa, to generate the competence scores and the diagnostic feedbacks. Conceptual analysis shows that the framework that is proposed here would have the ability to deliver superior assessment reliability in comparison with traditional approaches, which would be demonstrated through an example whereby the framework would be able to deliver 91.7% accuracy, F1-score of 0.90 and correlation of 0.88 with human experts’ assessments. The framework offers scalable language assessment, personalized feedbacks and detection of learners that need assistance.