LSTM for Sentiment Analysis

In this notebook, I implemented a recurrent neural network (Long short-term memory) using PyTorch that performs sentiment analysis.

Here I used a dataset of Amazon baby products reviews, accompanied by product names and rates. You can find it here

Network Architecture

The architecture for this network is shown below.

The layers are as follows:

1. An embedding layer that converts our word tokens (integers) into embeddings of a specific size.

2. An LSTM layer defined by a hidden_state size and number of layers

3. A fully-connected output layer that maps the LSTM layer outputs to a desired output_size

4. A sigmoid activation layer which turns all outputs into a value 0-1; return only the last sigmoid output as the output of this network.

It is not possible to push the model's state_dict here due to its size. If you need it, feel free to contact me.

Dataset

It's a CSV file consisting of reviews of Amazon baby products. You can download it from Kaggle. It consists of product names, reviews, and ratings associated with each. Bellow, you can see dataframe info:

Head of the data: