Exciting new Generative AI features coming soon! Stay tuned!
About
Fakebooks is a website created in order to apply Machine Learning and Deep Learning Techniques in a real world's dataset or problem.
General Information
This is a fun project!
This is a personal fun project created with limited (free) resources/tools.
- Without this project wouldn't have been possible. Besides of all the massive dataset available for the users, it also offers free GPU (30h a week), essential for training models using deep learning techniques.
- framework was used to create the website.
- Thanks to , the deployment of the website in Flask and the trained model (using PyTorch) is free and easy-peasy. This project uses the free version so there are limitations to it and makes the prediction a little bit slower.
- is my favorite machine learning framework 💗 so it was used to train all the models in this project.
- was used to design and create the background and kali (the bot).
- Bootstrap, using this perfect framework gave this simple but effective website design.
- Google Drive came quite handy for image hosting to expose my work in this website.
Book Genre Prediction
Already Available!
The machine learning trained model is able to identify the genre of the book by "reading" the blurb (also known as book summary or description). You can know about it all in my GitHub page where there is also available a document explaining the whole process. It includes different machine learning and deep learning models that were used and the benchmarks. CMU Book Summary Dataset was used to train the model. However, to balance the given data, summaries from goodreads were collected (web scraping) and also used for training. Right now it is trained to only detect the genres: Science Fiction, Crime Fiction, Non-Fiction, Children's Literature, Fantasy, Mystery, Suspense and Young Adult Literature. It has an accuracy of about 65%.
Book Summary Generator
In Progress!
Train a model to write summaries/plots in different genres. The user will only have to choose a genre and the trained model should be able to provide an unique summary/plot for the user. I've already worked in a similar project which you can check in my TV Script Generator notebook.
Style Transfer
In Progress!
Given a content image and another style image, the trained model is capable of creating a new target image containing the content and the style component of the original images. Learn more about it in Image Style Transfer Using Convolutional Neural Networks, by Gatys and also in my Style Transfer notebook.