Towards the Construction of an Inclusive and Fair Educational Environment

Education, undoubtedly, is one of the most fundamental means for people to gain personal and professional development. Given its importance, both researchers and practitioners have endeavored to apply various technologies to construct numerous educational systems and tools to facilitate teaching and learning in the past decades. For instance, with the development of Web technology, a large number of online learning platforms and web-based learning management systems have been developed and deployed for use, e.g., Khan Academy, Coursera, edX, Moodle, and Blackboard. In the meanwhile, it is worth noting that techniques like Data Science and Machine Learning (ML) have been increasingly applied in analyzing the unprecedented amount of data collected by these platforms and systems, e.g., using predictive analytics to identify at-risk students and providing early help to enable them to accomplish the study. Though numerous achievements have been made, the current application of these technologies still falls short in terms of the following two aspects. Firstly, very limited efforts have been made to enable more underprivileged students, e.g., those from developing countries or without a post-secondary degree, to receive a quality education and gain further development. Take MOOC platforms like Coursera and edX as an example. Though providing up to thousands of free courses, these platforms, more often than not, are helping students who are already highly-education (with a Bachelor’s degree or more) to upskill, while the underprivileged students are largely underrepresented. Secondly, the applied ML techniques tend to capture the human bias hidden in the data and deliver predictive outcomes that are potentially harmful to disadvantaged groups of students. For example, it has been reported that female students, compared to their male counterparts, were often predicted as being less likely to accomplish their study.

Therefore, this project aims to investigate issues that are essential to the construction of an inclusive and fair educational environment. In particular, there will be three lines of research in this project, which will, respectively, focus on developing techniques (1) to empower more underprivileged students to receive education; (2) to detect and eliminate human bias (i.e., instructors’ unconscious bias) towards the underprivileged students; and (3) to detect and eliminate machine bias (i.e., systemic and algorithmic bias) towards the underprivileged students. In the first line of research, we will focus on students whose disengagement with learning can be explained by their poor financial conditions and the subsequent limited amount of time they can dedicate to learning besides earning a living. Our solution is to enable these students to earn money by applying the knowledge they acquire from courses to solve paid tasks in online marketplaces (e.g., UpWork) so that they can support their learning. Potential research topics include how to effectively retrieve paid tasks that are relevant to a specific course from online marketplaces, what working mechanism should be introduced to enable students to successfully solve paid tasks, and what impacts will these real-world tasks have on students’ learning performance. In the second line of research, we will focus on applying ML techniques to automatically identify and eliminate instructors’ unconscious bias in their teaching practices (e.g., writing feedback). Potential research topics include how to systematically quantify instructors’ unconscious bias hidden in their written feedback, and to what extent such bias affects students’ learning performance, how language technologies can be adapted to eliminate such bias (e.g., providing machine-generated feedback-writing suggestions). In the third line of research, we will focus on developing not only accurate but also fair ML techniques, with the aid of which students of different protected attributes (e.g., gender and ethnicity) will receive equalized opportunities to enhance their learning performance. Potential research topics include how to accurately model learning performance across all groups of students, and how to generate fair course recommendations regardless of students’ backgrounds (e.g., recommending both STEM and non-STEM courses to female students).