News Quality Index (The Bullsniffer)

Helping to curb the current trend of misleading, manipulative and untrustworthy content

Comment

Inspiration

We live in a world where taking advantage of human psychology to misdirect and bait masses of people is extremely prominent. We also live in a world where people do not want to have authorities (such as governments) to manage how we live our lives. Through the chaos the selfish are winning as they are able to take advantage of the gap in the market where there is no consensus over how we should tackle their actions. Cambridge analytica and brexit are great recent examples of that.

We want to make it easy for people to identify poor quality and misleading content, but also create a standard around how content should be written. I actually take little interest in debate so I am increasingly upset that so much time in the world is spent debating over non-facts and as a society we are getting lost and overwhelmed. When a government presents their manifesto, or a news vendor writes about what someone said, how can we trust anything that's being said unless it's backed up by sources/data.

What it does

This is a browser extension that will let users know the quality of an article, based on multiple factors, without them having to read through it. It extracts the text from the article and uses NLP techniques to create an aggregated score for the quality of the article.

How I built it

Chrome extension based on create react app. It sends the browser text to an express API where I have used Tensorflow for measuring toxicity, and PerfectTense API for measuring the quality of writing.

Challenges I ran into

Of course there was a lot more I want to do especially as I arrived late (and am flying solo) but the task is huge anyway. Not only that but the decision around what is the right and wrong metric should be heavily scrutinised so I did not see much point in mapping that out in a definitive manner. A lot of bespoke libraries will also need to be written because anything that lacks transparency could be classed as containing bias. However the machine learning models themselves could also contain inherent bias.

Accomplishments that I'm proud of

Managed to get a lot done in a short time despite not turning out exactly as I'd hoped. It's an area that is not my speciality so I'm relatively pleased.

What I learned

Since graduating with a Physics degree around 6 years ago, going through different roles and now more recently contracting as a front-end developer it's been really interesting to look at a broad range of disciplines for this hack. I have now revisited machine learning and tried out some libraries that did not exist previously. I've also worked with browser extensions, which is something I had never had to do.

What's next for News Quality Index (The Bullsniffer)

We need to create bespoke algorithms, particularly to recognise factors like misleading headline and manipulative language.

I think it's crucial to have an open codebase that can be probed, critiqued and contributed to (subject to code review by a core team) in order to gain trust and consensus independent of political standing.

I want to add blockchain in order to get a consensus started for select third party 'authorities' to be able to vote on a simple metric like "Is this trustworthy"

As the popularity of this technology increases and as the algorithm improves we will have opportunities to apply the same logic for the index to low hanging fruit like a news website, fake review analysis, embed into social platforms, etc.

Built With

Share this project:

Updates