From Information to Insight: Leveraging LLMs for Open Aspect-Based Educational Summarization

This paper addresses the challenge of aspectbased summarization in education by introducing Reflective ASPect-based summarization (ReflectASP), a dataset that summarizes student reflections on STEM lectures. Despite the promising performance of large language models in general summarization, their application to nuanced, aspect-specific summaries in educational texts remains under-explored. Re-flectASP eases the exploration of open-aspectbased summarization (OABS), overcoming the limitations of current datasets and annotation complexities. We leverage GPT-4 for generating reference summaries and propose a selfrefine framework to enhance summary quality. Our work benchmarks the capabilities of different language models in this novel context, contributing a unique dataset and insights into effective summarization strategies for educational content. We will make our model, dataset, and all human evaluation results available at urlannonymized_for_review. methodological challenges while collecting manual 042 annotations on aspects and guiding annotators to ex-043 tract specific sentences from generic summaries for 044 aspect-based summaries. These challenges arise in 045 formulating reference summaries, making it diffi-046 cult to evaluate the generation quality of models in 047 an aspect-focused context. 048 This paper argues for domain-specific aspect 049 construction and appropriate evaluations, focus-050 ing on opinions in the educational domain in the 051 form of student reflections. Student reflections 052 provide valuable insights into students' learning 053