pytorch-sentiment-analysis

This repository is a set of four step-by-step tutorials that teach you how to build models that can read text and decide whether it carries a positive or negative tone. The specific task throughout is predicting the sentiment of movie reviews, which is a classic and well-understood problem that makes it easy to see whether a model is working.

Each tutorial is a self-contained notebook you can open directly in a browser and run without setting up anything on your own computer. The first tutorial introduces the simplest possible approach: a neural bag-of-words model, which treats a sentence as an unordered collection of words and learns which words tend to signal positive or negative meaning. The second adds a recurrent neural network, a type of model that reads words one at a time and tries to remember context as it goes, using a popular variant called LSTM.

The third tutorial switches to a convolutional approach, which scans over small windows of words rather than reading the whole sequence at once. The fourth and final tutorial loads a pre-trained BERT model, a large transformer that was trained on enormous amounts of text before being fine-tuned for this specific task. Each tutorial builds conceptually on the one before it, so reading them in order gives a clear progression from simple to sophisticated.

The tutorials are written for Python 3.9 and use PyTorch as the core framework. Dependencies install with a single command. An older set of tutorials exists in a legacy folder for anyone who worked through the previous version and wants to reference it, but the main four notebooks are the current recommended path.

This is a learning resource aimed at people who want to understand how text classification works in practice, not just in theory. It does not ship a production tool or a web interface. If you want to understand how modern language models approach simple classification tasks, this is a clear and hands-on starting point.