006_405 Found - Chatbot service for senior citizens

Inspiration

Inspired by the chatbot in Telegram and Siri, as well as the aging population in Asia, we wanted to come up with a tool that can enable effective communication with seniors to know their needs, help them if needed. 

What it does

This chatbot will focus on four major sections of conversations that are common in real life: greeting, entertaining, help and mental health. When the user initiates a conversation, we will try to provide relevant answers for the users. This would, psychologically, help the seniors. 

How we built it

Step 1: convert all English words into one-hot representation Step 2: Convert one-hot vectors into embedded word vectors Step 3: Parse the embedded word vectors into the encoder section Step 4: Generate the output word vectors in embedded format sequentially. Step 5: Output word vectors fed into the Dense Layer, converted back to one-hot representation Step 6: Identify the respective English words output

Challenges we ran into

There are not much NLP-related dataset on the internet, which puts us in the scenario where we have to generate the dataset (write English conversations) on ourselves Time is limited, so we do not have too much time to write a dataset that covers a broad spectrum of conversational topics Debugging skills and developing a workable model in a short period of time, which is very challenging

Accomplishments that we're proud of

The structure of the video and the functionality of the chatbot, especially when the input prompts contents relating to psychology and mental health which returns answers that are fulfilling. 

What we learned

How to convert our previous theoretical knowledge into a product that solves real-life challenges. This often requires a lot of fine-tuning and adaptation of our model to fit in to the particular challenge we are facing. In this project, we observe how the breadth and quality of our dataset (the English conversations that we write ourselves) will affect the generated sentence output. The reason that our chatbot is capable of returning fulfilling answers to psychological or mental health related contents is because we deliberately input the relevant English conversational topics into our dataset, which affects the final outcome.

What's next for 006_405 Found - Chatbot service for senior citizens

We will try to add more features on this chatbot: bring internet of things, Health +, multi-language accessibility and outreach. If they are successfully done, we would enlarge our target audience and therefore help more seniors in need. 

Built With

• keras
• lstm
• natural-language-processing
• python