A Controlled Synthetic Benchmark for Educational Aspect-Based Sentiment Analysis: corpus, human-validation study, and code

Artifacts for the paper A Controlled Synthetic Benchmark for Educational Aspect-Based Sentiment Analysis. Includes a 10,000-review synthetic educational ABSA corpus over a 20-aspect pedagogical schema, the Herath (2022) student-feedback corpus mapped to a 9-aspect overlap for external transfer, per-row LLM label-faithfulness audit scores, a three-rater human-annotation study validating the audit on the synthetic corpus (Fleiss kappa 0.70; human confirmation of declared aspects rises monotonically with the audit score), and the code to reproduce the benchmark, transfer, and filtering results. Best-per-target BERT checkpoints (synthetic-only transfer, synthetic-pretrain plus Herath fine-tune, and top-50%-faithfulness-filtered) ship in the companion archive course_absa_checkpoints.zip on this record.