The tutorial/walk-through on LSTM is here:
https://github.com/kunal-kotian/deep_learning_sentiment_classification/blob/master/model/sentiment_classifier.ipynb
The notebook above is intended to be a tutorial providing a walk-through and explanation of the steps required to build a variable length sequential input recurrent model in TensorFlow.
The tutorial covers:
- Minimal text preprocessing
- Using word embeddings (GloVe)
- TensorFlow?s Dataset API for batching and loading data
- Efficient handling of variable length input
The dataset used in the tutorial consists of Yelp restaurant reviews which are:
- Polarized (1 or 2 stars = negative; 5 stars = positive)
- 400 to 500 characters per review
All steps followed for preprocessing the dataset used in this tutorial can be found in the notebook at:
https://github.com/kunal-kotian/deep_learning_sentiment_classification/blob/master/filter/filtering_reviews.ipynb
Running the notebook at the link above will help you generate the file reviews_labeled.feather used in the sentiment_classifier notebook.