Image Processing unravels scientific reasoning procedure

Scientific argumentation is an essential practice through which scientific knowledge is generated, communicated and refined. By engaging in scientific argumentation, scientists generate evidence-based claims with theoretical backing, examine the constraints of their investigations and critique their own and others’ ideas. For students, scientific argumentation provides genuine opportunities to express and develop their understandings of topics in science by generating and selecting evidence, elaborating scientific reasoning and persuading their peers through rhetorical and dialogical discourse.

Currently, the role of visual representations in scientific argumentation remains poorly understood.Given the current trend of learning at scale accompanied by technological advancements (eg, electronic drawings, digital photos and three-dimensional models), students can produce plenty of images in every class. In this project, we investigated how visual representations support student scientific reasoning processes.Particularly, I explored approaches to quantify the associations between the visual features and the quality of students' scientific arguments.

Pei, B., Xing, W., Lee, H. S. (2019). Using Automatic Image Processing to Analyze Visual Artifacts Created by Students in Scientific Argumentation. British Journal of Educational Technology, 50(6),3391-3404

Pei, B., Xing, W., Zhu,G. & Antonyan, K. (2021). Visual Representations in Scientific Evidence-based Reasoning Processes. International Journal of Science Education(Under Review)

Machine Learning approaches detecting students' online learning patterns

AI algorithms are used in predicting and evaluating students’ learning performance based on large-scope heterogeneous learning data recorded by the learning platforms.We first situated the AI approaches such as natural language processing (NLP), machine learning in the online learning settings for predicting students’ learning performance explicitly using students’ forum discussion data, in combination of the theories related to social interactions, learning engagement, and self-determined motivation (SDT). The analytical results suggested it is possible to make inferences about students’ learning motivations and affective states using students’ posts in forum discussions. And the results also illustrated the effectiveness of these algorithms in extracting underlying patterns from large scope learning data to make predictions about students’ learning performance. The second aspect we worked on is identifying at-risk students as early as possible using AI algorithms, while providing interpretable insights for instructors to provide individualized interventions. Unlike many current approaches, we proposed a sequence-based framework explicitly capturing associations among students’ learning behaviors rather than simply aggregate the behaviors when making predictions about students’ learning statuses.

Pei, B. & Xing, W. (2021). An Interpretable Pipeline for Identifying At-Risk Students. Journal of Educational Computing Research. 07356331211038168.

Xing, W., Pei, B., Li, S., Chen, G. & Xie, C. (2019). Using learning analytics to support students’ engineering design: The angle of prediction. Interactive Learning Environments, 1-18

Tang, H., Xing, W., & Pei, B. (2019). Time Really Matters: Understanding the Temporal Dimension of Online Learning Using Educational Data Mining. Journal of Educational Computing Research, 57(6), 1326-1347.

Tang, H., Xing, W., & Pei, B. (2019). Beyond Positive and Negative Emotions: Looking into the Role of Achievement Emotions in Discussion Forums of MOOCs. The Internet and Higher Education, 43, 100690.

Tang, H., Xing, W., & Pei, B. (2018). Exploring the temporal dimension of forum participation in MOOCs. Distance Education, 39(3), 353-372.

Bibliometric analysis for multimodal learning analytics

Multimodal Learning Analytics (MMLA) has huge potential for extending the work beyond traditional learning analytics for the capabilities of leveraging multiple data modalities (e.g. physiological data, digital tracing data). To shed a light on its applications and academic development, a systematic bibliometric analysis was conducted in this paper. Specifically, we examine the following aspects: (1) Analyzing the yearly publication and citation trends since the year 2010; (2)Recognizing the most prolific countries, institutions, and authors in this field; (3) Identifying the collaborative patterns among countries, institutions, and authors, respectively; (4) Tracing the evolving procedure of the applied keywords and development of the research topics during the last decade. These analytic tasks were conducted on 194 carefully selected articles published since 2010. The analytical results revealed an increasing trend in the number of publications and citations, identified the prominent institutions and scholars with significant academic contributions to the area, and detected the topics (e.g. characterizing learning processes using multimodal data, implementing ubiquitous learning platforms) that received the most attention. Finally, we also highlighted the current research hotspots attempting to initiate potential interdisciplinary collaborations to promote further progress in the area of MMLA.

Pei, B. , Xing, W. & Wang, M.(2021). Academic development of multimodal learning analytics: a bibliometric analysis. Interactive Learning Environments, 1-19.

Fair AI to ensure education equity

While Machine Learning algorithms and technologies have the potential to facilitate teaching practices, there have been great challenges in real applications, including providing student-centered fair predictions and presenting transparent and explainable predictive processes. As such, the future direction of my research falls into two directions: (1) Exploring AI approaches to generate fair predictions among different students while ensuring predictive accuracy in educational settings: I will investigate approaches that can produce equitable inferences for students with different backgrounds, device metrics that can evaluate the fairness level of the predicted outcomes, and implement strategies to mitigate the detected unfairness in the predicted outcomes. With this research, I hope to propose a unified framework for examining the fairness of AI techniques across different student groups while solving high-stakes educational issues; (2) Implementing platforms for visualizing the decision-making processes of AI algorithms to increase the prediction transparency: Approaches for explaining machine learning models such as local interpretable model-agnostic explanations (LIME) and SHAP will be adopted and customized for explaining the predicted outcomes situated in the learning contexts. Data visualization techniques like Plotly-Dash and D3.js will be adopted to build interactive platforms to present the explanation process, ensuring end-users with limited AI backgrounds understand and challenge the decision-making processes.

Works In Progress Fair Intervetions: A Fairness-aware Framework for end-to-end Discrimination detection and mitigation in Educational settings