EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Concept-Based Tabulation and Summarization for Automated Note Making from Biology Textbooks

Nithyakalyani A

Journal of Intelligent Decision Making and Information Science · 01.10.2026 · 3 (3) · 93-124

DOI: 10.17762/jidmis.v3.5633 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Intelligent Decision Making and Information Science, online erschienen am 01.10.2026. Verfasst von Nithyakalyani A. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Automated note making can help students revise large volumes of textbook content, but existing summarizers produce running prose that does not isolate the key terms and definitions on which science learning depends, and existing text-to-table methods either require parallel training data or assume a fixed, date-ordered schema suited to history rather than science. This paper proposes an unsupervised concept-based tabulation and summarization framework that converts a biology textbook section into two linked notes: an extractive summary and a concept table whose rows are organised around key terms and whose cells are copied verbatim from the source. The framework identifies concepts with a cascade of definitional-pattern rules and a spaCy recognizer fine-tuned on glossary-derived silver labels, induces optional attribute columns from the section, and introduces concept-aware word-frequency ranking (CA-WF), which favours sentences that define or mention the table's concepts. For evaluation, OSB-Note (OpenStax Biology Notes) is constructed as an openly licensed benchmark of 204 sections from OpenStax Biology 2e that uses each section's author-written summary and glossary as references, together with an 80-section cross-book test set. On 83 held-out test sections, CA-WF achieved the highest ROUGE-1 recall (56.62%) and concept coverage (46.28%) among nine rankers; its coverage gain of 5.43 points over plain frequency ranking is significant against every ranker (Wilcoxon, Holm-corrected p < 0.05) at only 3.9 ms per section. In tabulation, definitional patterns gave the best term F1 (26.68%), while the cascade raised recall (31.18%) and row retention (15.60%), most clearly on the cross-book set (45.17% recall). An instruction-tuned language model (Qwen2.5-1.5B-Instruct) left 59.87% of its definitions and 85.75% of its summary sentences ungrounded in the source and was about 130 times slower, whereas every cell produced by the framework is traceable to a textbook sentence. The entropy-weighted Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) ranked TR-BM25 and LexRank first and CA-WF third under equal weights, identifying CA-WF as the preferred method when key-term coverage is the priority. Results on CNN/DailyMail show that the advantage of frequency ranking is specific to term-dense educational text.

Zitieren

APA

A, N. (2026). Concept-Based Tabulation and Summarization for Automated Note Making from Biology Textbooks. Journal of Intelligent Decision Making and Information Science, 3(3), 93-124. https://doi.org/10.17762/jidmis.v3.5633

Vancouver

A N. Concept-Based Tabulation and Summarization for Automated Note Making from Biology Textbooks. Journal of Intelligent Decision Making and Information Science. 2026;3(3):93-124. doi:10.17762/jidmis.v3.5633

BibTeX
@article{A2026,
  author = {A, Nithyakalyani},
  title = {Concept-Based Tabulation and Summarization for Automated Note Making from Biology Textbooks},
  journal = {Journal of Intelligent Decision Making and Information Science},
  year = {2026},
  volume = {3},
  number = {3},
  pages = {93--124},
  doi = {10.17762/jidmis.v3.5633}
}
Kostenlos starten →