-
Example of chat with AuthentiMate
-
NLP Test and Run
Inspiration
With social media such as Whatsapp becoming more popular as the main platform for communication, we see so many of our close friends and relatives fall victim to fake news and propaganda sent through messages. This has made us realise how influential social media can be and how important it is that we ensure the public can use messaging services in a safe environment.
What it does
AuthentiMate is a Whatsapp messaging service created with Twilio where users can forward their messages to check for fake news in one tap. From there, the message is processed and analysed before a reply is sent back to the user. This includes the reliability of the news as well as the influence of emotion in the way the news is written, hence indicating the likelihood of the news being real. This allows users not only to know if the message received contains unreliable or biased information but also to prevent propaganda from being spread and further igniting fear among others.
How we built it
To commence with, we started by choosing Whatsapp messaging instead of SMS. This is because this platform would allow us to reach out to more users and the intended effect was closer to our team goal. We tried to use Twilio to link to the Whatsapp sandbox and started by receiving simple commands from the bot. Next, we created a javascript server that can communicate with the Google Cloud AI. The AI was created by inserting 40,000 different sample.txt to train the bot and running the programme for hours. Finally, we worked hard to try to link the bot, AI and server together, which after several tries turned out successful. Finally, we finished the Whatsapp messaging bot by connecting it to the GoogleCloud AutoML API.
Challenges we ran into
Most of our teammates had little to no prior experience in programming. Hence, we had to work with what we had in the limited time and use the skills we were learning for the first time. As we began, we ran into the issue of being unable to connect the server to the Google Natural Language cloud platform. Luckily, after countless debugging and researches on javascript, we managed to resolve it. Then, we struggled with controlling the output produced by the bot. Once again, our cooperative team effort helped us resolve it, in that we created JSON keys to log in to the Google Cloud in js.
Accomplishments that we're proud of
We are extremely proud of being able to create functional parts of this project despite the challenges and having absolutely no prior experience in most of the skills needed to tackle this project. We are very proud of the effort we put in and being able to train this model on data that we cleaned and processed ourselves. Despite not having high expectations, we managed to successfully create this bot which is a huge achievement for us. Our team couldn't have been more grateful for being able to work on this project.
What we learned
Besides learning so much about programming than we have ever known before, we also discovered a lot about the Twilio communication platform by attending the workshop provided. We also learned about programming fake news identification using NLP and how to use cloud services like the ones we used with GoogleCloud. We were a team with limited experience but we went for an ambitious project and learned a lot along the way.
What's next for AuthentiMate
From this point onwards, we would need to continue to train our models using the latest data as sometimes it has blind spots regarding current topics of interest. It would also be amazing if we could be able to expand it into various platforms and eradicate fake news not only in social media platforms but also including news websites by using the fake news detector as a Chrome extension.
Log in or sign up for Devpost to join the conversation.