-
As you can see, it's very much a work in progress, however it's able to identify many clickbait-y links
-
However, the plugin leaves unoffending websites mostly untouched.
Inspiration
Although clickbait blockers already exist, they are often times blocking content from clickbait producing websites, and not clickbait in general. This means that articles published by otherwise reputable news sources that utilize clickbait-y titles will slip through the cracks. Fortunately, machine learning is the perfect simple solution, as it can analyze the content without relying on hard-coded blacklists.
What it does
Click Exterminate finds all links with clickbait-y titles and removes them from the web page. Users can manually train the model by pointing out clickbait and non-clickbait, which means that the plugin only gets better with age.
How I built it
The chrome plugin is fairly standard, as it only really relies on jQuery and a few minor libraries. There's a Node.js which the clickbait identification model resides on, but it doesn't do much more than that. Initially, I was using Google's cloud prediction API, which worked wonderfully, but was slower than native categorization and API limits made its use impossible in this project. I later transitioned to a Node.js library which performs Naïve Bayes categorization, which performed spectacularly and had a decent enough accuracy.
Challenges I ran into
It took me 18 hours to decide on a project, meaning I was on a rather tight schedule. Additionally, I had gone through a large bit of development before I hit Google's call limit, so the switch to a Naïve Bayes categorizer was quite late. Additionally, false positives were and still remain a concern. Due to the nature of the algorithm, it's likely that false positives will be minimized over time, but it's still inconvenient for the demo. Finally, figuring out the scope at which to remove the offending content was a bit ambiguous, so I went for the conservative approach of simply removing the link.
Accomplishments that I'm proud of
This is my first time using machine learning, so I'm rather thrilled that it worked so well. However, I'm also quite happy with the skills I learned while trying to come up with this idea, such as experimenting with eye tracking, dabbling in image recognition, and
What I learned
I learned a great deal while making Click Exterminate. Aside from simply learning some basic text analysis techniques with machine learning, I had also learned that machine learning can be accessible, even to a beginner with a limited amount of time. Additionally, I had learned a great deal about the HTML 5 speech recognition API (though that was used in another, failed project).
What's next for Click Exterminate
If Click Exterminate were to be released at some point in the future, many improvements would have to be made. By adding users, the data and model are sure to improve at an increasing rate, however, I would like to look into other text classification algorithms. I also believe that quite a bit of time would need to be spent in the plugin its self, as it doesn't usually eliminate every trace of the clickbait article. Finally, the whole thing just needs some general stability improvements.
Log in or sign up for Devpost to join the conversation.