Word Embeddings for a Disciplinary Tutoring System

Question Answering (QA) systems have made remarkable progress in information retrieval techniques, especially in their ability to naturally access knowledge resources by querying and retrieving correct answers to various questions. In tutoring, these systems can help by reducing the requirement for interaction between learners and tutors and allowing learners to post their queries and receive answers for the same. Hence, we propose a disciplinary tutoring system based on a domain ontology ONTO-TDM (ontology for teaching domain modeling) and natural language processing (NLP) techniques to facilitate access to information and answer the learners' questions. Recently, deep learning algorithms have achieved impressive success in various natural language processing tasks. The basic concept of these techniques is to compute a distributed representation of words from continuous vectors, also known as word embedding. In this study, we use deep learning-based word embedding models for a disciplinary tutoring system. Our goal through this work is to find out whether word embedding could significantly improve the response generation task of the suggested system. Therefore, we have built word embeddings using the word2vec skip-gram model with different training parameters on a large corpus composed of question-answer pairs. Experimental results show that using the word2vec model has a significant impact on the accuracy of the proposed tool.