Effectiveness of Recurrent Layers and IndoBERT Regularization for Virtual Assistant Intent Classification on the Indonesian MASSIVE Dataset — Reproducibility Code

Abstract

Google Colab notebook and original results (state.pkl) that reproduce Tables 2 and 4–10 and Figures 3–4 of the article "Effectiveness of Recurrent Layers and IndoBERT Regularization for Virtual Assistant Intent Classification on the Indonesian MASSIVE Dataset". Eight models are compared on the Indonesian subset of MASSIVE 1.1: TF-IDF + SVM, BiLSTM, IndoBERT, IndoBERT-BiLSTM, IndoBERT-BiGRU, and IndoBERT with R-Drop, Supervised Contrastive Learning (SCL), and both combined, each trained with three random seeds and compared using McNemar's test with Holm correction. Uploading original_results/state.pkl to Colab regenerates all tables and figures without retraining; see README.md for details.