TV-Script-Generator
Using LSTMs to generate a TV Script (Keras).
You can go through my blog for a detail insight on this project:
Introduction
In this project, I am generating a TV Script using LSTMs, which is a type of Recurrent neural network. Our training data is a small subset of Simpsons Dataset. Simpson is a TV series having 27 seasons. The dataset contains the conversation between various characters. I will be using Keras framework in this tutorial.
This is how my pipeline looks like:
- Preprocess the data.
- Create a model for training.
- Create a model for inference(generate new script).
- Generate the script using the inference model.
- Post process the script.
Results
How to furthur improve
We can generate a much more realistic script by doing a few things such as: using the large dataset since our dataset is a small subset of the original dataset, tuning the hyperparameters, experimenting with different network architectures like bidirectional and encoder-decoder, applying text augmentation etc.