Mirrur

Hack RPI Fall 2019 x Project Mirrur Assisted Journalling with NLP for Mental Health

Wolfram Award Winner 🎉

Goal

Help pre-depressive/hopeless behaviors through customized supportive messaging and, if necessary, provide a report to share with a mental health care professional to find inflection points.

We collected natural language from anonymous text messages and respond with context-sensitive texts per their hopelessness index (0-1). At lower levels, we referred users to mental health specialists with an online anonymous report they could access. This report offers the ability to find changes in hopelessness over time.

Presentation Link https://docs.google.com/presentation/d/1ld6yPPlV6nb9XL2jQKcTsk6SC9VIfFV61sRsD5DCPRI/edit#slide=id.p4

NLP Heuristics (Averaged for hopelessness index)

• Absolutism/Polarization Index (0-1 Index for usage of words like "always", "never", "completely")
• Global Sentiment (0-1 Positivity Word Values Summed and mapped to 0-1 score)
• Last Entry Sentiment (Same as Global Sentiment but just for latest entry to weight the present)

Note: Polarization was found to be highly predictive of hopeless behavior, for this reason, we made sure to focus our efforts on here. It was a challenge weighting the index based on the amount of these determiners, adjectives and adverbs relative to the total count of these parts of speech in the body of text

Tools & Technologies

• Python NLTK Package for Parts of Speech Tagging (Polarization Index)
• Google Cloud Platform Natural Language API (Sentiment Analysis)
• Twilio SMS API
• DigitalOcean (VM & Networking)
• .xyz domain (https://mirrur.xyz) [will most likely be down later]

Pictures/Screenshots

Example Journal

Example Full Report

Contributors

• Brian Wu
• Andres Orbe
• Justin Kwong
• Jacob Shomstein