<aside> 🧠

NLP CLASSIFIER is a multilingual NLP research project for English, Tagalog, and Cebuano text, combining lightweight language routing, topic modeling, and transformer-based zero-shot classification.

</aside>

1. Project overview

The project investigates NLP behavior on multilingual and code-switched text rather than assuming English-only data.

It combines several techniques because language identification, topic discovery, and semantic classification are different tasks:

• a fast statistical language detector;
• LDA as a traditional topic/classification baseline;
• BERTopic for embedding-based topic discovery;
• BART-Large-MNLI for zero-shot semantic classification;
• FastAPI for serving inference.

The goal is not to claim that one model solves all Filipino multilingual NLP. The project compares approaches and records where each one is useful or weak.

2. Problem

Tagalog, Cebuano, English, and mixed-language text create challenges for pipelines designed around clean monolingual corpora.

Short messages can contain:

• English technical terms mixed with local-language grammar;
• multiple languages in one sentence;
• sparse context;
• spelling/orthographic variation;
• topics that are difficult to infer from word frequency alone.

The project therefore separates inexpensive language routing from deeper semantic analysis.