We are living in a crisis. We have been forced by an invisible enemy to keep social distancing, not knowing how this can impact our lives. This has weakened us so it is wonted that we keep watching over our shoulders, afraid of our own shadows, not knowing what to believe. Out of desperation, we can follow last and unthinkable ideas. Therefore, it is vital to control the spread of fake news, in order to control anxiety and desperation in our community. This uproar of misinformation, highly spread across multiple platforms, can shape our views and leave people more skeptical so, when the times comes to embrace legitimate political or health care quality measures, the community will struggle and take some time to act rationally in a period where time is precious.
Do you remember the news articles you read in the last hour? Can you count them all? We are not aware of how much exposure to news we are currently experiencing and how legitimate is all the information we so blindly follow. Therefore, it is essential to have control and authenticate all the news in order to contain any possible mental health issues that have been hauting us.
What it does
Our solution is based on a browser extension that acts as a personal self-control assistant. It will alert users if they are seeing a lot of news during the day. This is an important resource for providing feedback to users.
In addition, it is essential to pay attention to the uproar of disinformation, spread across various platforms. A report from a reliable source provides real data, but many reports exist only for social alarmism. Thus, it is essential to score the news by legitimacy and topic. For that, we put scores on all news, providing instant feedback to the user.
Imagine that users believe they are in the presence of news that is not well rated or that is not relevant to them, in this case, they can “blame” the articles and the machine learning algorithm will adapt to the topic and content and adjust the scores. In this way, users take an active role in controlling the news!
How I built it
To build the Backend, we use Python3 and Flask.
For the Machine Learning algorithm we also used Python3 for development, Google Colab for running on GPU and Kaggle for datasets.
In this section, we wanted to explore how artificial intelligence technologies, specially machine learning and natural language processing, might be leveraged to combat the fake news pandemic. Our model uses very common and very powerful techniques to assess the available features composed by words carefully positioned in order to form coherent sentences.
Essentially, the proposal method to detect fake news is designed in two main parts – preprocessing and data preparations in which every title and news content are embedded in vectors in an orthogonal space; and the second part where the model is trained in order to capture different associations and underlying meanings.
Since we are dealing with an imbalanced domain due to the high number of positive cases compared to negative ones, our model was evaluated based on the AUC performance measure. The following graph plots our results.
Challenges I ran into
Our biggest obstacle was the response times, both in the machine learning algorithm and in the server, so that there was not much delay in the extension. Another concern we had was the comunication between the server, responsible for managing our application, and the Google Colob notebook tied to the machine learning implementation.
Accomplishments that I'm proud of
With n3wstr4ck, users have a personal assistant for their mental health. This reduces the exposure to fake and unfounded news and gives instant feedback to the user. It also helps to control users' mental health by providing self-control tools through an assistant that can be configured according to preferences. It will alert users if they are exceeding the advised number of seen news in a period of time.
What I learned
We would like to thank EUvsVirus for the opportunity to work in the context of this matter, which is a challenge for all of us. There are a lot of people concerned about COVID in fields that sometimes we are familiar with and have tools which can detect hidden patterns and provide reliable insights and fast measures to stop this spread.
This was the first hackaton we did with 4 people and it is not easy to reach the end and have a practical result. There are things that can be improved, but our work during the 48 hours is remarkable and we are happy with what we accomplished.
What's next for n3wstr4ck
If you want a personal assistant for your mental health, install n3wstr4ck in your browser! We will try to seek funding for the application to reach millions of people. For this, we will open donation schemes, as in Ad-Block and similar services, and allow to show personalized advertisements across different platforms such as social media.