A Controlled Synthetic Benchmark for Educational Aspect-Based Sentiment Analysis

2026-05-25

AI Summary

This study addresses the scarcity of aspect-annotated student feedback in education, which hinders sentiment analysis research aimed at course improvement. To bridge this gap, the authors construct a controlled synthetic benchmark dataset comprising 10,000 human-generated reviews covering 20 teaching-related aspects. Through three iterative rounds of prompt refinement and expert review, the realism of the synthetic comments is significantly enhanced. This dataset constitutes the first structured, reproducible synthetic benchmark for aspect-based sentiment analysis (ABSA) in education, incorporating label-aware sampling, fine-grained attribute control, and realism optimization to mitigate the paucity of real annotated data. Experimental results demonstrate that BERT-based models achieve a micro-F1 score of 0.2930 on the held-out test set and attain a transfer performance of 0.4593 on an external real-world subset, validating the efficacy of the synthetic data.

Abstract

Educational aspect-based sentiment analysis (ABSA) can support course improvement, but public aspect-labeled student feedback remains scarce because educational reviews are private, institution-specific, and expensive to annotate. This study introduces a controlled synthetic benchmark for educational ABSA built from 10,000 synthetic course reviews with explicit train-validation-test splits and a 20-aspect pedagogical schema spanning instructional quality, assessment and course management, learning demand, learning environment, and engagement. The corpus is generated with sampled target labels, sampled nuance attributes, and a realism-tuned prompt refined through a three-cycle judge-editor procedure. On the resulting benchmark, local baselines with TF-IDF, two-step transformers, and joint encoders show that the task is nontrivial; the strongest untuned model, BERT, reaches a held-out detection micro-F1 of 0.2760, while a modest lower-rate BERT schedule improves this to 0.2930. Full-test GPT-based inference with gpt-5.2 reaches 0.2519 micro-F1 in zero-shot mode and 0.2501 with retrieval-based few-shot prompting, placing batch inference above the classical baseline and close to the compact joint encoders. A conservative external evaluation on 2,829 mapped student-feedback reviews from Herath et al. yields a micro-F1 of 0.4593 for BERT on a 9-aspect overlap, indicating partial synthetic-to-real transfer. Realism and faithfulness analyses are reported as generator diagnostics that clarify how the benchmark was stabilized and where label noise remains. The study therefore contributes a synthetic educational ABSA corpus, a documented generation procedure, and a reproducible benchmark setting for a domain in which public labeled data remain difficult to obtain.

Problem

• educational aspect-based sentiment analysis
• synthetic benchmark
• student feedback
• aspect-labeled data
• data scarcity

Innovation

• synthetic benchmark
• educational aspect-based sentiment analysis
• controlled text generation
• realism-tuned prompting
• transfer evaluation