YouTube SafeGuard

Prototype 1
A picture of how it would look like from a user's perspective.

Inspiration

Our team has always been concerned about how YouTube is handling its comment moderation. Our friends have experienced many nasty hate comments, and we recognize the impact these online spaces have on our daily lives as technology becomes more entangled with our social experiences. We want to help make these challenges easier to overcome by providing better toolsets for people to view their comments, either by filtering out Spam comments or being able to view the right type of comments at the right time so you would not have to expect a gross comment every time you view the comment section.

What it does

Our application takes in the comments from a YouTube video, then run them through Co:here's NLP models to categorize them as Spam comments, Positive comments, Neutral comments, and Negative comments. Afterward, you can then view these comments from our Chrome extension at your convenience.

How we built it

We divided this challenge into 3 parts. Part 1 is taking in the comments from YouTube and turn them into parsable data, part 2 is finding the right data for Co:here's NLP to learn from, and part 3 is bringing these data back to our users. We aim to ensure these processes are as intuitive and as inclusive as possible by having a smooth transition between the parts, and providing the right samples for the NLP model through a sufficient amount of samples from various sources, including different languages.

Challenges we ran into

One of our team's hardest challenges was learning how to integrate YouTube's structure into our code. Different videos have multiple formats and it was a challenge to generalize comment replies for extraction.

Another noteworthy challenge is to find the right samples for the NLP model. We recognize that there may be cultural biases that NLP models have while testing the limits of the model, so we have to be careful to correctly label the many comments while going through different comment sections from different backgrounds to ensure that our model does not mislabel new inputs and harm the user experience.

Accomplishments that we're proud of

One of our crowning accomplishments is that our strong teamwork has made the process of developing an intuitive web application much simpler and that we have overcome many of the inherent challenges of learning new toolkits and frameworks to integrate them into a working process that will contribute to improving the quality of life.