A browser extension that utilizes sentiment analysis to find and highlight constructive comments on various social media platforms that oppose the users worldview in order to encourage them to break out of the echo chambers the internet has allowed us to construct.
This is an ideal set of steps we would take. What we focus on and when things are completed is subject to change.
March and April
[ ] Fix up the repo
[ ] Collect data, particularly social media data.
[ ] Read up on the latest NLP breakthroughs such as BERT, Transformers, etc.
[ ] Read up on some of the specific sub-problems such as text summarization and topic classification
May
[ ] Develop an effective political leaning classifier
[ ] Research methods of incorporating ML into web extensions, and how they should be structured to ensure they aren't resource intensive for the user
[ ] Develop an effective topic classifier
[ ] Create a dead simple testing platform and test the effectiveness of the combined leaning/topic models.
June
[ ] Develop an event classifier and determine the general feasibility of this segment of the project, as it's subject to external factors
[ ] Layout a framework for the extension or application, determine server requirements
[ ] Continue testing real world performance of existing classifiers using local testing platform
[ ] Build up a prototype extension with the existing models for very basic functionality
July
[ ] Ideally, soft launch on the MVP, though this is likely not feasible
[ ] Develop a set of text summarizers using a variety of parameters, data subsets and techniques if necessary
[ ] Continue developing extension. I'm going to learn to hate web programming all over again this summer
[ ] Establish the framework for developing the fake news classifier. Owing to the potential politicized subjectivity of what counts as fake news, this is an important step before development for the credibility of the project
Roadmap
This is an ideal set of steps we would take. What we focus on and when things are completed is subject to change.
March and April
May
[ ] Create a dead simple testing platform and test the effectiveness of the combined leaning/topic models.
June
[ ] Build up a prototype extension with the existing models for very basic functionality
July
[ ] Establish the framework for developing the fake news classifier. Owing to the potential politicized subjectivity of what counts as fake news, this is an important step before development for the credibility of the project
August