Spam Detector

Comparative analysis of text classification methods: fastText, BERT, RAG approach, and scikit-learn models on a single corpus.

Models

Model	Approach	Key Features
FastText	Shallow neural network	Fast training, subword embeddings
Scikit-learn	TF-IDF + Logistic Regression	Lightweight, interpretable
BERT	Transformer	High accuracy, contextual embeddings
RAG	TF-IDF + kNN retrieval	No training required, explainable
VectorDB RAG	ChromaDB + LaBSE	Semantic search, multilingual

Installation

uv sync # select required groups to work with

Usage

See experiments/ directory for training and evaluation scripts.

Testing

uv run pytest

Hugging Face

Published models and datasets are available in the spam-detection collection.