While awareness and resources devoted to mental health have greatly increased in recent years, detecting mental health issues still remains as the greatest problem. In most cases, it is difficult for friends and family to discern when someone is suffering from a hidden condition. Tragically, people are often unaware of the subtle clues that only make up the tip of the iceberg. Our team decided to rectify this issue and provide a tool to detect general negativity and more specifically what kind of issue is the cause. We felt that many issues go unnoticed either because we are not paying enough attention or we simply just are not attuned to the small clues here and there. We decided that an assistant that analyzed the text of your messages would greatly assist in covering up our cognitive blind spots.
What it does
Our demo application allows for a string input to be sent to our Google cloud server, after which it will be classified with our trained model to identify the distributions of mental health issues that may underlie the supplied phrase along with a general negativity score.
How we built it
We scraped data off of many mental health oriented subreddits on Reddit.com in order to train a model to recognize different stressors. We then used Google’s cloud services’ Natural Language Processing to determine general negativity and also Google’s AutoML to train a model to detect 6 commonly occurring issues relating to mental health: relationships, depression, suicide, isolation, suicide, isolation, self-harm, and abuse. We trained the model with sentences from Reddit posts, resulting in a model we were later able to use towards our larger project. We integrated the model into a web application, accepting text input and measuring the proportional representation of each of the 6 categories.
Challenges we ran into
• Multiple issues with the android api to send and receive data • Issues scraping data from the internet - reddit api request caps • Working with new technology in order to finish our projects
Accomplishments that we're proud of
• Multiple issues with the android api to send and receive data • Successfully scraping meaningful data from the internet • Persevering and learning new technology in order to finish our projects
What we learned
• The different stressors in mental health and the statistics about each one • How to use some of Google's suite of cloud products and what is important in a dataset • Effective organizational structure of a project
What's next for PsychSMS
Our intent is to implement this into an SMS messaging application that will use our trained model, finding trends based off of the messages, analyzing if the messages are alarming or not and predicting what possible issues the user may be facing and what steps could be used to help the user.