TweetWatch

• 

  The reason something needs to be done.

• 

  Screenshot of a sample run on the UI.

Slideshow

TweetWatch Presentation

Inspiration

We wanted to address a dire lack of awareness and attention towards the most deeply troubled in our society.

What it does

TweetWatch helps identify individuals who may pose a risk to themselves or others by analyzing their posting behavior on the popular social media app Twitter.

How we built it

We used a fine-tuned DistilBERT model in conjunction with publicly available tweet data from user profiles to assign ratings and positivity and negativity to their recent tweets.

Challenges we ran into

We didn't have a budget for API access so we had to work around not being able to access the Twitter API directly or the HuggingFace APIs without limit.

Accomplishments that we're proud of

One success we found while testing was that if a given user added new tweets that were clearly problematic, their risk factor increased on the next run without fail.

What we learned

We learned that machine learning has a lot to offer the psychological and therapeutic communities,

What's next for TweetWatch

Training the neural net with more data, also creating a more aesthetic and more accessible UX.

Built With

• bert
• css
• git
• github
• html
• huggingface
• python