DESIGNING AI TEACHER ASSISTANT ON ONLINE-COURSE BASED ON WORD2VEC TECHNOLOGY

• Main Authors: Pavel Aleksandrovich Rozhkin, Igor Nikolaevich Nekhaev, Kirill Anatol’evich Markin
• Format: Article
• Language: English
• Published: Science and Innovation Center Publishing House 2018-05-01
• Series: International Journal of Advanced Studies
• Subjects: сопроождение обучения на онлайн-курсе, технология word2vec; векторизация вопросов; векторное пространство текстов; классификация тематики вопроса; поиск релевантных ответов
• Online Access: http://journal-s.org/index.php/ijas/article/view/10725

The purpose of this work is to develop an AI teacher assistant, who can find answers to online course participants questions among answers previously published at the training forum. Currently, there are already successful experiments on the use of artificial intelligence systems (IBM WATSON) in online training. In this paper, we investigate the possibility of constructing such a system using word2vec technology. A two-stage method for finding an answer to a question is constructed. Method use word2vec technology for vector representation of questions and answers. At the first stage, the subject matter of the issue is determined and, if it corresponds to the theme of the forum, then the articles most relevant to the question are searched. A real situation was simulated with 16 themes and 80 answers to possible questions within the section of the online course “Linear Algebra and Geometry”. The question-answer system was designed and its performance was evaluated. The parameters have been chosen to achieve the best result. In 83% of the cases, the relevant answer to the formulated question was contained among the top 3 responses that the system offered. The issues of further development of applied approaches and increasing utility of the constructed question-answer system are considered. Purpose: developing an AI teacher assistant, who can find answers to online course participants questions among answers previously published at the training forum. Methodology: vectorization of questions and answers, neural network classification of the subject matter, construction of the answers rating. Results: acceptable accuracy in finding a relevant answer to a question are received. Practical implications: The results of the research can be used as a basis for designing an AI teacher assistant in online courses.