A Browser Extension for Civil Discussions
#mozsprint 2017 Interview Series
Tyler (@TyTheSciGuy) is a grad student at McGill University with an interest in human bias, fake news, and barriers to communication. I met Tyler at our recent Working Open Workshop in Montreal. Since then, he’s been combining technology and psychology to overcome human bias in his project Echoburst.
I interviewed Tyler to learn more about EchoBurst and how you can contribute June 1–2 at #mozsprint.
What is EchoBurst?
EchoBurst is a browser extension that will use natural language processing techniques to evaluate the comments and articles you read online and make several judgements regarding its content. First, it identifies what the topic of the comment or article is, then determines the position taken by the commenter, and lastly evaluates if it might be considered ‘toxic’ or likely to shut down discussion. It then compares this information with your position on the relevant topic, and if it is non toxic and disagrees with your stated beliefs, it will prompt you to read it. In order to encourage use, it will also measure the variety of opinions you’re spending time on, and score you on the diversity of your media diet (the broader the range, the better).
Why did you start EchoBurst?
I’ve always been disheartened by the disconnect between science and policy, and the inability to turn evidence into action. One of the big causes of this is misinformed voters, and this issue has come into the spotlight lately under the names “fake news” and “political polarization”. Finding common ground is the only antidote to the poison of polarization. EchoBurst was started to make it easier to take the first step of finding that common ground, which is engaging in discussion with people you don’t agree with. We also hope it will make it easier for people to find more diverse sources of news and information and in general diversify how people get their information. The goal isn’t to change opinions in one direction or the other, it is to broaden the content that informs those opinions.
How did you come up with this project?
Several friends and I started discussing the phenomenon of echo chambers, and the enormous effort that it takes to engage with opinions outside our chosen spheres. Reading things that challenge our views is hard enough on it’s own, and we were frustrated with the additional effort needed to first track down those opinions. To combat this, we envisioned a website or social media platform, where the site algorithm would work counter to how many work right now, where it would make things you tended to click on less likely to appear, forcing you to branch out. However, we realized an extension would be a more passive way for our tool to operate without requiring additional steps like logging into a specific site, which might discourage use.
What problems have you run into while working on this project?
There will certainly be other problems to come, but the largest so far and by far has been a lack of data. What we’re trying to do is pretty specific, and requires a vast amount of labelled data. We originally intended to read and label comments manually and directly in order to create a dataset, but we realized that we need far more data than that approach can provide. So instead we are hoping to pre classify data based on the source, and scraping the entire blog or section of the site under the assumption that the topic and opinion are relatively consistent.
The largest problem we foresee long term is a lack of engagement due to the inherently uncomfortable nature of having your views challenged. We are working hard on developing incentive schemes and reward systems, but there’s no way to directly overcome human nature, or this wouldn’t be a problem in the first place. We are hoping to at least lower the sense of being threatened by external views to a point that even those who don’t set out with the intent of challenging their own viewpoint may find themselves using it.
What kind of skills do I need to help you?
You don’t need any particular skills to help as there are many tasks that need to be completed that just require judgement and a willingness to discuss and ask questions when needed. Even if you just want to contribute to the discussion, or have opinions on how you think we could most effectively market or attract users, that’s very valuable.
That said, there are certain skill sets that are particularly suited to this project. Right now the main skills we’re looking for are:
- Machine learning and natural language processing expertise, especially regarding sentiment analysis
- Front-end development experience to help get the app packaged properly for use
- Insight into ways of making our data collection faster and more accurate, as our current approach makes some assumptions (though we do feel they are justified)
- Expertise in human behaviour and reward systems to help structure the app to encourage use.
How can others join your project at #mozsprint 2017?
I’ll be available on June 1st and 2nd to answer any questions, either through issues on the EchoBurst GitHub repo or the related Gitter chat . If you have questions or want to start contributing before then, there is a CONTRIBUTING.md page, as well as an issue for new contributors to ask questions. The main tasks that we’ll need help with during mozsprint will be the labelling of blogs based on sentiment and topic. Discussions on various topics such as developing an incentivization scheme will also be a focus. We may also be looking to have people labeling specific comments for our testing set (the blogs are for the corpus and training). Anyone who wants to work on something more specialized or technical is free to open an issue or ask questions on the existing Contributing issue so we can coordinate what you want to do to make sure two people aren’t working on the same thing.