Depression has become increasingly common among teenagers – especially teen girls, who are now almost three times as likely as teen boys to have had recent experiences with depression. One-in-five teenage girls – or nearly 2.4 million – had experienced at least one major depressive episode over the past year in 2017. By comparison, 7% of teenage boys (or 845,000) had at least one major depressive episode in the past 12 months. The total number of teenagers who recently experienced depression increased by 59% between 2007 and 2017. The rate of growth was faster for teen girls (66%) than for boys (44%). Graph depicting the rise in depression


1. Emotional changes

  • Feelings of sadness, which can include crying spells for no apparent reason
  • Frustration or feelings of anger, even over small matters
  • Feeling hopeless or empty
  • Irritable or annoyed mood
  • Loss of interest or pleasure in usual activities
  • Loss of interest in, or conflict with, family and friends
  • Low self-esteem
  • Feelings of worthlessness or guilt
  • Fixation on past failures or exaggerated self-blame or self-criticism
  • Extreme sensitivity to rejection or failure, and the need for excessive reassurance
  • Trouble thinking, concentrating, making decisions, and remembering things
  • Ongoing sense that life and the future are grim and bleak
  • Frequent thoughts of death, dying, or suicide

2. Behavioral changes

  • Tiredness and loss of energy
  • Insomnia or sleeping too much
  • Changes in appetite — decreased appetite and weight loss, or increased cravings for food and weight gain
  • Use of alcohol or drugs
  • Agitation or restlessness — for example, pacing, hand-wringing or an inability to sit still
  • Slowed thinking, speaking or body movements
  • Frequent complaints of unexplained body aches and headaches, which may include frequent visits to the school nurse
  • Social isolation
  • Poor school performance or frequent absences from school
  • Less attention to personal hygiene or appearance
  • Angry outbursts, disruptive or risky behavior, or other acting-out behaviors
  • Self-harm — for example, cutting, burning, or excessive piercing or tattooing
  • Making a suicide plan or a suicide attempt

What it does

It is very hard for youth in third-world countries to open up about depression or their mental state. So we created an algorithm that can analyze if the user is depressed/suicidal or on the verge of depression using his/her day-to-day chat conversations. The algorithm after analysis suggests the user remedies to lighten/brighten up the latter’s mood and prevent depression. The algo has multilingual support, which means you can analyze the mood of a french/Japanese person too. The algorithm is very easy to use and can be embedded in any chat app using Graphql API. Moreover, no user data is stored anywhere on the server. The data is only used to understand the user’s sentiment and suggest the latter remedies accordingly.

How we built it

  • Django Server: Cloud server based on python to process all the input chat data and suggest remedies accordingly.
  • Language Corpus: We use NLTK, iNLTK, and Google databases for different language corpora.
  • Heroku: To deploy the server container.
  • Google Cloud: To deploy the website.
  • Javascript: Interface to showcase how it works.
  • GraphQL: To build a stable API for public release.
  • Google PostgreSQL: Database service to temporarily store users' chat records and analyze the data.

Accomplishments that we're proud of

  • The mood analysis and remedy suggestion algorithm is very accurate with an accuracy score of 87%.
  • The algorithm works with many languages like English, Spanish, Hindi, Japanese, French etc.
  • The Graphql API is very stable, efficient and can be used in any project or chat app.

Challenges we ran into and what we learned

  • It was very difficult for us to process languages other than English as the language corpora data is very limited for other languages.
  • We learned a lot about Natural Language Processing and Sentiment Analysis in this project. It was our first time working with multi-language support.
  • We used and studied many sentiment analysis models classified on the basis of their accuracy.

What's next for iHear

  • We wish to collaborate with many doctors/therapists to build a more accurate analysis model and bring better suggestions to users.

How to test

  1. Go to the website
  2. Try chatting from the perspective of both users
  3. You will get suggestions on top of the chat about the user's mood and what to talk about to lighten up his/her mood


Share this project: