Christina Krist, Paul Hur, Kevin Hall, Chris Palaguachi
Naturalistic video recordings are an important data source for studying learning, especially when learning is considered an emergent outcome of a complex interactional social-material-epistemic system. Video recordings allow researchers to analyze the dynamics of talk and interaction. However, current methodologies require Herculean efforts to conduct analyses that simultaneously attend to complexity and nuance at scale, especially those that focus on the auditory and visual-spatial aspects of the video recordings.
This workshop will introduce existing AI-based techniques for extracting audio and visual-spatial features from data. It will also introduce a framework for how to integrate those techniques into existing qualitative workflows with an aim of “layering on” additional information to prioritize rich description and theoretically informed judgements over typical AI aims of automation and efficiency. After an introductory talk, each breakout session will focus on introducing and applying one of three AI-based techniques to classroom video data: (a) extracting body pose and movement coordinates using OpenPose; (b) detecting voice activity and extracting prosodic features using OpenSmile; and (c) transcribing verbal utterances and using natural language processing using a range of open-source tools.
Multimodal collaboration analytics: Investigating how group processes influence persistence and well-being in STEM education:
Morten Misfeldt, Daniel Spikol, Liv Nøhr, Viktor Holm-Janas
Recent developments in collaboration analytics use multimodal data that utilize sensors, video, audio, and human observations to understand how small groups of learners interact. One key aspect is the multimodal nature of data, and the need to combine different data sources to establish a valid foundation for analysis.
The workshop will focus on the process “datafication” in a groupwork setting and use a wearable badge (currently in prototype stage) that combines computer vision, a QR code and audio collection to understand the patterns of interaction, conversation, and type of learning activity. By “datafying” the interactions between the students, we aim to identify critical indicators in the groupwork related to reflection, persistence, and a positive approach to failure. In our methodology, the data from the badges is combined with a survey tool that measures non-cognitive aspects (e.g. motivation, self-efficacy and attitude to science), as well as observations and video recordings.
We will touch upon the balance between ethical and data-lightweight applications of multimodal learning analytics, and the need for solid information about student activities and discourse. Using a mock-up example, we will introduce tools for data representation and analysis, and discuss different examples of data. We will then facilitate discussions on what datatypes that can be collected, and how they can be analyzed using multimodal strategies.
Our focus will be on (1) upscaling (from small qualitative examples with dense data, to large samples of lightweight data), (2) on joint analysis of different strands of data, (3) and on ethical and practical concerns we meet when working with these data.
GrAIND-MASER: Generative AI for Non-Data Scientists in MASER
Giora Alexandron, Alexandra I. Cristea, Elad Yacobson, Yael Feldman-Maggor, Abigail Gurin-Schleifer, Nadav Kavalerchik
Educational data mining (EDM) involves the application of data mining techniques within the educational domain to extract valuable insights from educational datasets. By leveraging statistical analysis, machine learning algorithms, and data visualization, EDM seeks to uncover patterns and trends, with the purpose of improving pedagogical design and supporting data-driven decision-making. Generative AI (GAI) refers to the branch of AI that focuses on creating new content, such as images, text, or even music, that mimics the style and characteristics of the original training data. Large Language Models (LLMs) are a specific type of generative AI that excel in generating coherent and contextually relevant text. LLMs, with ChatGPT as the chief example that will be used throughout this workshop, are trained on vast amounts of text and can produce human-like responses, engage in conversations, and assist in various language-related tasks. LLMs can assist in educational data mining by providing language-related capabilities that support data analysis, interpretation, and communication.
In the workshop, we will concentrate on the potential of LLMs as a research methodology that can analyze student and teacher texts.
The workshop will have the following stages:
- Stage 1: practice basic work with the chat through the web interface for interactive, beginner-level prompt engineering (in English).
- Stage 2: ChatGPT+ simple programming: We will perform basic prompt engineering, evaluate results and debug.
- Stage 3: prompt engineering with ‘fine tuning’ and ‘context’ (e.g., requesting the chat to score a response + providing a rubric and an example of a response scored according to that rubric)
- Stage 4: Integration of ChatGPT with other tools
Discussion and recap
Optimal Learning Moments: Measuring Academic, Social, and Emotional Learning in Daily Life:
Barbara Schneider, Lydia Bradford, I-Chien Chen
Measures of social and emotional wellbeing conventionally are tested in laboratory studies. However, whether reviewing work on mindset or grit, the results often overlook the conditions and contexts in which the participants’ responses are measured. The context in which emotions occur is influenced by many factors, including previous feelings, time of day, your relationship with those around you, and your feelings towards the activity or task you are undertaking. More importantly, your engagement with the tasks at hand is strongly related to whether the learning experience is interesting, you have the knowledge, skills, and cultural understanding regarding the learning task, and the task is challenging, leading you to try and figure out an answer that you don’t know. This is where our work is developing a new perspective on the fluidity of emotionality and its impact on your academic performance and wellbeing. This includes recent research by Bradford, which explores model specification when trying to understand the mediating effect of engagement on science achievement, providing insight into important methodological considerations when using digital repeated measures in-situ of student emotionality, such as how modeling the latent repeated measures directly (with structural equation modeling; SEM) compares to using student means over time (with a hierarchical linear model; HLM).
During the workshop, there will be a hands-on activity where groups of participants will be instructed to develop their own research design on emotionality using the Experience Sampling Method (ESM) and present it to their peers. Following this, real-time ESM results will be presented and explored at the end of the workshop to promote additional discussion on its use.
Winds of change in MASER: How will the future change MASER and how will MASER change the future - Roundtables & Panel
The final and closing session of our conference will focus on the future of methodologies in science education. Our main aim will be to discuss the vision our conference’s participants hold for the future of science education research methodologies, science education in general and the way the two visions affect each other. To achieve this we invite our attendees to participate in a slightly different conference session which will be divided into two consecutive seats.
During the first seat we will hold a round table discussion, involving several small and intimate discussion groups. The discussion groups will try to formulate statements and questions about the future of science education research methodologies and about the way these might change the way science is taught and learned.
During the second seat we will hold a panel of experts which will express their point of view regarding the statements and questions presented by the discussion groups. Panelists will be asked to discuss the issues raised by the discussion groups and answer their questions, while presenting their own views on the future of science education in general and MASER in particular.
