AI-enabled Chatbots are ubiquitous in both business and daily lives. They are not only providing buyers with instant support but also extending their assistance to the elderly. Our client has taken AI technology up a notch to help thousands of Japanese family members to stay in touch and take care of their parents.
- ML-enabled chatbot to exchange conversations with elderlies
- Connect to mobile application to update the log activities and exchange messages with family members
- Conditional Random Field for pattern recognition and structured predictions
- GPT2 for text summarization and text output generation
- Team size: 4 members
- Development time: 2.5 months
The client is among the big players in the Energy sector in Japan. It has been investing in the technology area to diversify its portfolio and offer more added services to the existing clients. One of the new projects that the company is planning to do is a robot assistant designed particularly for the elderly. People at old age usually stay home all day while other family members go to work or school. They may need a companion to make small talk with and remind them to take the medicines on time.
Therefore, instead of complex and highly integrated smart bots like Amazon’s Alexa or Apple’s Siri, a preprogrammed chatbot will serve them just right. In addition to chatting with the elderly, the robot can also help them call or send text messages to their friends and children. With that said, the company has decided to launch a Proof of Concept (POC) project to validate the initiative. It collaborates with GEM to build the chatbot from scratch.
To create the new chat application, the GEM team received only a small dataset of conversations between the Japanese team and senior citizens on a number of topics. Thus, the data to feed the machine was not sufficient to ensure accurate outcomes.
GEM team had put the collected data from the conversations through a pre-processing phase using NLP techniques such as filtering out stopwords, stemming, etc. GEM team had created similar datasets from the given input to train the chatbot to address the problem. This was to make sure that we have a considerable sample size to achieve possible accurate results. We applied the Conditional Random Field (CRF) method to classify the query’s context and created relations between any number of keywords. We also used GPT-2 to generate text and build question-answer systems.
It took GEM team two and a half months to finish building the chatbot application. We obtained more than 80% accuracy in answering the user queries using the chat application. The POC project was a pass, and the company was ready to put the application into production. The robot is expected to be in the market in the next few years.