1 The United Graduate School of Education, Tokyo Gakugei University (JAPAN)
2 Center for Information and Communications Technology, Tokyo Gakugei University (JAPAN)
About this paper:
Appears in: ICERI2023 Proceedings
Publication year: 2023
Pages: 8867-8874
ISBN: 978-84-09-55942-8
ISSN: 2340-1095
doi: 10.21125/iceri.2023.2258
Conference name: 16th annual International Conference of Education, Research and Innovation
Dates: 13-15 November, 2023
Location: Seville, Spain
It is important for students to reflect on their learning by engaging in their learning proactively to acquire competencies in each subject and the entire curriculum. Activities to reflect on learning are often carried out after classes or learning activities. At this time, students often reflect on learning and describe what they have engaged in, what they have realized, and how they should proceed with their learning. It is thought that rereading these written reflections while categorizing them, such as at the end of the learning unit or end of the semester, will provide an opportunity for summative reflection, i.e., reflecting on their learning process and outcomes and how they will proceed with their learning on the basis of what they have learned thus far.

However, it is not easy for students to categorize their accumulated written reflections by similar written reflections. It is also not easy for students to determine what of their learning process or outcomes each group of written reflections represents even if students can summarize or categorize their accumulated written reflections. Students tend to focus on rereading the text of written reflections.

The use of large language models has been attracting attention. Large language models are natural-language-processing models constructed by learning a large amount of text data and can perform various tasks related to text, including generating text. For example, by using large language models, we can obtain embeddings of text data, which is a vector list of floating numbers, for natural language processing, and cluster text data on the basis of the embeddings. We can also generate the commonalities of multiple text data as text.

By using large language models, if we can cluster written reflections of students and provide students the results of clustering and cluster names that represent the commonalities of written reflections belonging to the constructed clusters, such as “descriptions indicate about your growth and experience in a project”, we will be able to support summative reflection.

The purpose of this study was to support summative reflection of students by using written reflections. Specifically, we developed a summative-reflection support method that involves clustering for classifying text data by similar text data and generating cluster names that represent the commonalities of clusters of text data by using large language models.

With our proposed method, we first obtain embeddings for each accumulated written reflection using a large language model. We then carry out clustering and construct several clusters with similar written reflections. Next, we generate a name for each cluster that represents the commonalities of written reflections belonging to the constructed clusters. Finally, we provide students the clustering results and each cluster’s name. This will enable students to carry out summative reflection.

As results of simulation on clustering written reflections and generating each cluster’s name that represents the commonalities of written reflections belonging to the constructed clusters, we found that written reflections belonging to a cluster are similar to each other. We also found that the characteristics and commonalities of written reflections belonging to a cluster are generally expressed.

For future work, we will conduct detailed verification of generating cluster names for developing a system that automates our proposed method.
Summative Reflection, Written Reflection, Large Language Models, embeddings, Clustering, Cluster Name Generation.