At least 25 percent of people who are impacted by natural disasters could be diagnosed with the symptoms of post-traumatic stress disorder. Other researches also point out that victims of natural disasters typically experience fear of losing control on overwhelming emotions, and becoming mentally ill. Thus we found that providing more convenient therapeutic solutions with trustworthy AIs may has a huge market potential.

What it does

SASRA will provide online therapeutic solutions such as debriefing with chatbots, relaxing puppy pictures,etc. for people who suffered post traumatic stress or mildly stressed people with cutting edge AI technologies.

How we built it

For emotion detection, we trained a machine learning model from scratch on over 500 video files tagged with the emotions "sadness", "anger", "joy", and "fear". To do this, we parse each video files into 9 distinct images taken at even time intervals. These then get regrouped into groups of 3 frames with 9 channels each. After passing each grouping through individual CNNs, the resultants get passed through a GRU to capture time dependence, which then gets passed through Linear Layers to predict the emotion. This allows our app to take in a video file of the user's face to get their current emotional state. Then, it will direct the user to a relevant html page full of helpful recommendations and exercises to help improve their current mental health. Additionally, Watson Assistant is incorporated as a chatbot to facilitate more authentic therapeutic experiences.

Challenges we ran into

The connections between back-end python files and front-end web application; Time; Normalizing Video Data; Keep minds calm.

Accomplishments that we're proud of

Successfully trained the sentiment analysis model from scratch; Finished prototype of the web application.

What we learned

One can permute a matrix to efficiently format a tensor; Time flies.

What's next for Sentiment Analysis Stress Relieving AI (SASRA)

Finish the connections between back-end and front-end; Add more features to enrich user experience.

Built With

Share this project: