indonesian-nlp

Transformer Encoder for Indonesian intent classification — implementing Positional Encoding & Multi-Head Self-Attention from scratch, compared against fine-tuned BERT via 5-fold cross-validation. Includes overfitting investigation, early stopping, and FastAPI serving.

GPT-2 text generation lab: training a code-completion model from scratch (CodeParrot) and serving it via FastAPI, plus fine-tuning a pretrained Indonesian GPT-2. Results are documented honestly, including undertraining and overfitting findings — not just the wins.

IdSarcasm: Benchmarking and Evaluating Language Models for Indonesian Sarcasm Detection (Suhartono et al., 2024)

A Python NLP project to detect social media hate speech in Indonesian language.

A curated list of NLP resources for Bahasa Indonesia and Indonesia's regional languages

Indonesian Text-to-SQL benchmark for LLM evaluation with 30 scenarios across two open databases.

A simple FAQ search engine built with TF-IDF and Sastrawi (Indonesian stemming). Includes an audit that found Euclidean distance let an off-topic question nearly pass as a match — fixed by switching to cosine similarity.

Intent classification on Indonesian text using TF-IDF + Logistic Regression. 100% accuracy confirmed via 5-fold cross-validation, then stress-tested with edge cases that reveal what that accuracy actually does (and doesn't) mean.

Quantization research for compiler-verified LLM systems. Three arms: (A) Phase D IntLLM 1.58-bit ternary, 3 gates PASS; (B) v3.1 adaptive KV cache quant, 7/9 cells; (C) Phase E bilingual ID+EN, corpus v1.0 25.67B tokens, E2.4 honest negative result. In-kernel via FajarOS Nova. Apache 2.0. Made in Indonesia.

IndoELECTRA: Pre-Trained Language Model for Indonesian Language Understanding

Research code for our ICIMCIS 2026 paper: distant supervision with heuristic-augmented denoising for crypto NER on Indonesian Twitter using IndoBERTweet-CRF

Indonesian SMS spam classifier — SVM, Logistic Regression, and XGBoost compared with hyperparameter tuning.

Indonesian Gojek review sentiment analysis comparing TF-IDF Logistic Regression, SVM, and a dense neural network.

Multi-label hate speech & abusive language detection for Indonesian tweets using deep learning

A machine learning system for classifying Indonesian text sentiment (positive/negative/neutral) using XGBoost (TF-IDF) and transformer-based NLP, trained on 30K MyPertamina app reviews.

SentiRank is a thesis-stage decision-support dashboard for Spotify review analysis using IndoBERT sentiment inference, SVM aspect classification, and AHP/Fuzzy AHP ranking with a Next.js frontend and FastAPI microservices.

Social media analytics course materials: text mining, topic modelling, data collection (Twitter/X API & Twint), stance detection, user profiling (gender & occupation), and network analysis with PageRank.

NLP dashboard that turns TikTok comments about Indonesia's Makan Bergizi Gratis (MBG) program into sentiment, public issues, and risk signals.

☕ An AI-powered tool to understand customer sentiment for Indonesian coffee shops. Built with Python, FastAPI, and TensorFlow, and deployed automatically with GitHub Actions.

Proyek Machine Learning untuk menganalisis sentimen (Klasifikasi) dan memprediksi skor (Regresi) ulasan aplikasi "Wondr by BNI" menggunakan metodologi CRISP-DM, diimplementasikan sebagai aplikasi web interaktif dengan Streamlit.