This page collects several ideas of possible projects to help you figure out what might be your project to work on.
- Encoder/decoder transformer-based chatbot
- This huggingface/pytorch-pretrained-BERT repository contains several Transformer-based models that you can use.
- HRED-based sequence-to-sequence architecture
- There is a PyTorch implmentation of HRED
- A chat-bot with personification
- Emotions and emojis 🙈
- Dialog State Tracking Challenge
- Visual Dialog Challenge
- Visual Question Answering
- Generate discussions on current affairs (like two or more agents posting on a forum)
- You can start with this Web-page
Choose a project you would really like to work on but did not have time! Remember, anything* can be framed as a question answering task, and, furthermore, as a dialog!
*According to Socher
Below is a list of datasets that you can use in your projects:
https://docs.google.com/document/d/1QVVX0YV5_ebH5M9XUtT7VveD0v2wVQy69DeKkWqhcz4/edit?usp=drivesdk
The tutorial sessions will show you how to do basic things, such as loading data, constructing a model, training and testing. Unless you are an advanced learner already, it may be a good idea to expand the code from the tutorials for your project.
The dataset can be huge, and our time is limited. In early development, it may be wise to use a small subset of your training data. Do not make each development cycle hours long.
An innovative project may have an innovative task, or an innovative model, or both. Something is better than nothing. After you make things work, you will have more ideas and you can always add things on top of it.