LECTOR: Summarizing E-book Reading Content for Personalized Student Support

TL;DR

This paper introduces LECTOR, a model that contextualizes reading content from lecture slides to complement traditional reading-activity data. By extracting slide-topic relationships through an attention-based framework and integrating them with reading logs, LECTOR delivers topic-aware features that improve keyphrase extraction and aid in predicting at-risk students. Two experiments demonstrate: (i) superior keyphrase extraction over NLP baselines and a TF-IDF baseline, with strong human agreement, and (ii) modest but consistent improvements in at-risk prediction when topic preferences replace or augment traditional reading metrics. The approach enhances interpretability and supports personalized interventions, with potential extensions to multimodal content and time-aware analyses to further support educators and learners.

Abstract

Educational e-book platforms provide valuable information to teachers and researchers through two main sources: reading activity data and reading content data. While reading activity data is commonly used to analyze learning strategies and predict low-performing students, reading content data is often overlooked in these analyses. To address this gap, this study proposes LECTOR (Lecture slides and Topic Relationships), a model that summarizes information from reading content in a format that can be easily integrated with reading activity data. Our first experiment compared LECTOR to representative Natural Language Processing (NLP) models in extracting key information from 2,255 lecture slides, showing an average improvement of 5% in F1-score. These results were further validated through a human evaluation involving 28 students, which showed an average improvement of 21% in F1-score over a model predominantly used in current educational tools. Our second experiment compared reading preferences extracted by LECTOR with traditional reading activity data in predicting low-performing students using 600,712 logs from 218 students. The results showed a tendency to improve the predictive performance by integrating LECTOR. Finally, we proposed examples showing the potential application of the reading preferences extracted by LECTOR in designing personalized interventions for students.

Figures (16)

• Figure 1: Our proposed multimodal integration multiplies the Slide-based reading characteristics by a Matrix M of Slide-Topic relationships to obtain Topic-based reading characteristics.
• Figure 2: The methodology proposed by wang_2022 can be reformulated as a data transformation using a binary matrix of Slide-Topic relationships.
• Figure 3: The architecture of the Transformer encoder, including the Input sequence, the Positional encoding, the Self-attention mechanism, and the generated set of Embeddings.
• Figure 4: Example of the self-attention mechanism: The Self-Attention matrix quantifies the importance of a Key for a given Query, which is used to calculate a weighted average of the Value.
• Figure 5: Example of the cross-attention mechanism: The Cross-Attention matrix quantifies the importance of a Key for a given Query, which is used to calculate a weighted average of the Value.