The indications that someone may suffer from a mental illness are often not recognized due to their loved ones lack of experience recognizing warning signs or even self denial. By providing this service, we hope to foster an open discussion between parents and children about ways one can detect mental health issues as well as how to get help.

What it does

Torch is a Chrome extension that would run an algorithm on search queries in real time in order to detect signs that a person may be considering harmful behaviors or exhibit other symptoms of mental illness. It would then send a text message to an appropriate emergency contact regarding flagged search queries.

How we built it

After obtaining the search queries through Javascript, we used the Microsoft Text Analytics Sentiment Analysis API to first filter out queries that returned a positive score (queries that sounded positive or indicated a good mood). Negative queries would then be passed onto our machine learning algorithm (logistic regression), which we trained to differentiate generally negative comments from ones that possibly indicated an underlying mental disorder. We then used the Twilio SMS/MMS API to send a notification to a phone number containing the search query, and the owner of the phone number would then respond to the text message if this search query was something they were concerned about or not. This would then train and refine our machine learning algorithm.

Challenges we ran into

We had to learn how to integrate the Javascript and Python code into one cohesive application that was running in the background and thus lacked a user interface. We also had a limited amount of time and data to train our machine differentiating generally negative reactions from those that posed a mental health concern.

Accomplishments that we're proud of

This is the first time any of us had worked with machine learning, and everyone learned a new language (Javascript, Python, or both) in order to contribute to this project. Furthermore, it was the first hack that two of our members worked on and submitted.

What we learned

We learned how to implement machine learning algorithms and how to use a server to integrate Javascript and Python. Additionally, every single member learned a new coding language.

What's next

We plan to expand this idea to other forms of media, for example, in group messages. We also hope to design a multi-user web platform that would display data in a clearer format, and to move the server to the cloud so that the service would be accessible to more people.

Special Thanks

We'd like to thank all the organizers and mentors of LA Hacks and the sponsoring companies. Without them, we would probably still be stuck debugging our code! We'd also like to give a special shoutout to Matt Reyes, one of our mentors, for contributing much of his time helping us develop our machine learning algorithm.

Share this project: