We want to create more engaging stories for children to read. We believe that in children's entertainment, there is a gap between the world of fantasy and the real world. Our goal is to bridge that gap by involving the reader and their surroundings (by using location data) into the world of fantasy in the stories they read.
What it does
We use a deep recurrent neural network to train on a corpus consisting of various stories with multiple characters and plot arcs. Specifically, for this project, we trained on Aesop's Fables and characters from the Marvel universe. We also gather information about the reader, such as their name and location, to personalize the stories we generate. For instance, we incorporate landmarks and street names around the reader's location to make the story more interesting. In addition, we also use location data to gather data about the weather and use this to customize the story even more.
How we built it
We adapted a deep recurrent neural network (with LSTM layers) in Keras to train on our corpus and then generated the output text by performing an element-by-element prediction which would involve characters (including the reader), plot lines from the world of fantasy and information specific to the user's location.
Challenges we ran into
Training takes a lot of time and data. The primary challenge was dealing with limited computing power. However, we believe that with greater resources, this would work even better than it does in its current state.
Accomplishments that we're proud of
We're proud of coming up with a product that can generate custom-made stories that make some sense given the constraints.
What we learned
We learned that we need a LOT of data to get a great system working.
What's next for Wordsmith
So much. In addition to improving the features which are already present in the product, we also want to incorporate other aspects of the children's life into the stories such as their hobbies and interests (this may be a bit creepy though).
Log in or sign up for Devpost to join the conversation.