A Domain Specific Students' Assistance System For The Provision Of Instructional Feedback

An important strategy for the reduction of academic-related stress during online academic engagements is the integration of conversational agents that deliver real-time instructional feedback. While instructional feedback is either focused on structuring a learner's learning interaction style or providing clues when a learner is experiencing difficulties while engaging with a given task, a major challenge in the development process of conversational agents is the creation of interactions that are nearly indistinguishable from that of a human instructor's feedback. In this work, we take advantage of recent improvements to large language models and information retrieval by using bidirectional encoder representations from transformers to both decode and classify the intent of a message input and give feedback or recommendations that are relevant to the context of the message input. Compared to the baseline, we achieved a precision of 80% and an accuracy of 85%, which justifies our strategy of using the BERT model for our conversational agent for the integration of instructional feedback.