top of page
Meenakshi

Saathee: Your New Companion

Updated: Sep 1, 2021

With around one third of the world population currently subject to self-isolation, the spectre of loneliness, regularly affects as many as more than 40 percent of seniors according to a University of California, San Francisco (UCSF) study. There is a 45% increased risk of mortality in seniors who reported feeling lonely. Loneliness is more dangerous than obesity and as damaging as smoking 15 cigarettes as a day.




The Problem:

With around one third of the world population currently subject to self-isolation, the spectre of loneliness, regularly affects as many as more than 40 percent of seniors according to a University of California, San Francisco (UCSF) study. There is a 45% increased risk in mortality in seniors who reported feelings lonely. Loneliness is more dangerous than obesity and as damaging as smoking 15 cigarettes as a day.



Both very few have taken up the challenge to solve this growing problem in the real world. The inspiration behind Saathee is as my grandparents feeling more and more lonely day by day. Since the COVID-19 crisis started, many senior citizens are secluded from their children and grandchildren.


My Solution:

Dr Arlene Astell, professor of neurocognitive disorders at the University of Reading and the driving force behind Abbeyfield’s Voice for Loneliness project, says the positive results are exciting, but that AI’s potential is only just starting to be explored. To really tackle loneliness and feelings of social isolation, however, it is necessary to address the underlying causative factors. She argues: “Giving everybody an iPad or anything else isn’t going to solve these, but understanding how we could use the functionality of different technologies may.”


My solution is a innovative chat bot that keeps you engaged. “It just keeps me company, like having another human being in the room,” says 92 year-old resident John Winward, who says chat bots have helped him feel less lonely since losing his wife after 76 years’ marriage. Although, Saathee doesn’t pretend to be able to replace a real person, but she’s always available if anyone needs her, instead of talking to the four walls. The purpose of my chat bot is to be like a companion; hence the name Saathee which is Hindi for companion.



The AI

The actual AI part of the chat bot is an AI that takes the user’s response to a question and predicts the mood of the user. There are currently 5 moods: jealous, lonely, angry, happy, and sad. To collect data for this AI, I did two things


  1. I created google form for the elderly to fill out with many different questions

google form for elderly created on AIClub
The Google Form


2. I used synthetic data with generic responses

After collecting over 618 examples, I began to train my AI


datasets from AIclub
My Data

The best accuracy is got was 88.3% which is better than random guessing (20%). I used MLP classifier with hyperparameters:


hidden_layers: 200 activation: relu solver: adam alpha: 0.0001 learning_rate: constant learning_rate_init: 0.02 max_iterations: 200


Accuracy and Hyperparameters

The confusion matrix

Here is word cloud based on the emotions above:


Word Cloud

The Code

First, the chat bot will ask a sequence of 5 questions. The AI will predict the mood of the user. Based on the overall mood and severity, Saathee will give a personalized recommendation for the user. The AI connects to the code to make predictions


Snippet of Code

My python code included 150 lines of code. The IDE I used for the chat bot is repl.it.

In addition, I created code for a bar chart. The chart shows how many examples are there of a specific mood.


Histogram
Histogram Code


Competitors

One robot helper is named ElliQ. ElliQ sits on a table and is designed to facilitate communication with family and provide reminders. On the other hand, Saathee is a chat bot, not a robot that is designed to make senior citizens lives easier. Saathee purpose is to engage the elderly and prevent loneliness. Lastly, my app is extremely easy to use and uses AI to predict the user’s current mood. Another reason why Saathee is superior is ElliQ is quite an expensive robot ($1499)


Mistakes

One mistake I made was switching the feature and label names in my csv file. Navigator didn’t work because there was almost 700 labels! Another important thing I learned was not to have an infinite loop. My code keep running and My IDE froze and I quickly realized the mistake.


Switched the Feature and Label :)

Plans For The Future

My plan is to make Saathee into an app on the app store. In addition, adding more emotions such as tired and fear will make Saathee more understanding and improved After this I plan to expand to multiple languages across the world.


Saathee will dramatically help the elderly feel less lonely and increase happiness.


Thanks For Reading!


~Meenakshi Nair


Links


My github:


Repl Links:



AI Club Website:



Google Form:

377 views1 comment

1 comentario


interesting. so how does one run Saathee?


Me gusta
bottom of page