Analysis of Students' Attempts Trajectories in Learning Programming
Saved in:
| Title: | Analysis of Students' Attempts Trajectories in Learning Programming |
|---|---|
| Language: | English |
| Authors: | Idir Saïdi, Nicolas Durand, Frédéric Flouvat |
| Source: | International Educational Data Mining Society. 2025. |
| Availability: | International Educational Data Mining Society. e-mail: admin@educationaldatamining.org; Web site: https://educationaldatamining.org/conferences/ |
| Peer Reviewed: | Y |
| Page Count: | 11 |
| Publication Date: | 2025 |
| Document Type: | Speeches/Meeting Papers Reports - Research |
| Descriptors: | Programming, Online Courses, Visual Aids, Algorithms, Data Use, Learning, Foreign Countries, Natural Language Processing, Progress Monitoring |
| Geographic Terms: | New Caledonia, Ireland (Dublin) |
| Abstract: | The aim of this paper is to provide tools to teachers for monitoring student work and understanding practices in order to help student and possibly adapt exercises in the future. In the context of an online programming learning platform, we propose to study the attempts (i.e., submitted programs) of the students for each exercise by using trajectory visualisation and clustering. To track the progress of students while performing exercises, we build numerical representations (embeddings) of their programs, generate the trajectories of these attempts (i.e., the sequence of their attempts) and provide an intuitive visualization of them. The advantage of these representations is to capture syntactic and semantic information that can be used to identify similar practices. In order to describe these practices, we perform a clustering of these attempts and generate a description of each cluster based on the common instructions of the underlying programs. By studying a student's trajectory for an exercise, the teacher can detect if the student is in difficulty and help him. Our approach can also highlight atypical solutions such as alternative solutions or unwanted solutions. In the experiments, we study the impact of using embeddings to identify common practices on two real datasets. We also present a comparison of different dimension reduction methods (PCA, t-SNE, and PaCMAP) for the purpose of visualization. The experimental results show that code embeddings improve results compared to a classical approach, and that PCA and t-SNE are the most suitable for visualization. [For the complete proceedings, see ED675583.] |
| Abstractor: | As Provided |
| Entry Date: | 2025 |
| Accession Number: | ED675685 |
| Database: | ERIC |
| Abstract: | The aim of this paper is to provide tools to teachers for monitoring student work and understanding practices in order to help student and possibly adapt exercises in the future. In the context of an online programming learning platform, we propose to study the attempts (i.e., submitted programs) of the students for each exercise by using trajectory visualisation and clustering. To track the progress of students while performing exercises, we build numerical representations (embeddings) of their programs, generate the trajectories of these attempts (i.e., the sequence of their attempts) and provide an intuitive visualization of them. The advantage of these representations is to capture syntactic and semantic information that can be used to identify similar practices. In order to describe these practices, we perform a clustering of these attempts and generate a description of each cluster based on the common instructions of the underlying programs. By studying a student's trajectory for an exercise, the teacher can detect if the student is in difficulty and help him. Our approach can also highlight atypical solutions such as alternative solutions or unwanted solutions. In the experiments, we study the impact of using embeddings to identify common practices on two real datasets. We also present a comparison of different dimension reduction methods (PCA, t-SNE, and PaCMAP) for the purpose of visualization. The experimental results show that code embeddings improve results compared to a classical approach, and that PCA and t-SNE are the most suitable for visualization. [For the complete proceedings, see ED675583.] |
|---|
