Automated Educational Content Summarization Using TextRank Algorithm in E-Learning Management Systems

The accelerating volume of digital educational content hosted on institutional learning management systems (LMS) has produced a structural mismatch between content provision and content engagement: Moodle's 2024 global usage analytics report documents a mean course reading completion rate of 48% across 330 million active users—a figure that operationalises the cognitive overload challenge of modern LMS-hosted instruction and establishes the magnitude of the content reduction problem that automated summarisation addresses. TextRank (Mihalcea and Tarau, 2004), a graph-based extractive summarisation algorithm that applies PageRank random-walk scoring to a sentence similarity graph constructed from the source document, offers a uniquely appropriate solution for institutional LMS deployment: it requires no training data, operates without external API dependencies, processes documents in 0.84 seconds at zero marginal cost, and produces summaries whose pedagogical coherence is rated 4.12 out of 5 by domain expert evaluators—within 0.06 points of GPT-4's 4.18 despite a 5.71-fold speed advantage and USD 0.042 per-document cost saving. This study evaluates TextRank against four comparison methods (LSA, LexRank, BERT-Ext, GPT-4) across a corpus of 12,840 educational documents drawn from six LMS platforms (Canvas: 3,840; Moodle: 3,200; Blackboard: 2,400; Google Classroom: 1,600; Microsoft Teams Education: 1,000; Schoology: 800), spanning three academic disciplines (STEM: 4,840 documents; Humanities: 3,640; Business: 4,360) and four document types (lecture slides: 4,280; textbook chapters: 3,640; course notes: 2,840; video transcripts: 2,080). The TextRank configuration employs a TF-IDF cosine similarity sentence graph, damping factor d = 0.85, convergence threshold$\varepsilon=0.0001$, sentence window size 4, and a 28.4% target compression ratio validated through pilot evaluation on 1,284 held-out documents. TextRank achieves ROUGE-1 = 0.484, ROUGE-2 = 0.248, and ROUGE-L = 0.412 on the evaluation subset-below BERT-Ext (ROUGE-1: 0.512) and GPT-4 (0.528) but superior on every operational deployment dimension. Deployment across the six LMS platforms serving 24,800 students over 36 months demonstrates a 24.5 percentage point improvement in content engagement rate (40.9% to 65.4%) and an 18.4% reduction in mean study time per learning objective, with student satisfaction of 4.24/5 and instructor satisfaction of 4.08/5.

Hali tarjima qilinmagan