• 

  Custom Logo For Project

Inspiration

One of our team members is an active user of Reddit and he noticed that misinformation was being spread on Reddit from time to time and it was hard to differentiate the facts from opinions due to the limitations of text only posts. We saw that we could benefit society by creating an app that helps people pick out the fake news from the real and useful resources on Reddit.

What it does & How we built it

It a website that (using Reddit's API), reads in the titles of posts on a specific subreddit and analyzes the sentiment(with the Python library Textblob) and classifies it as a true or fake story all based on the title alone(with a feed forward neural network made with Keras/Tensorflow and ported over to the website for inferencing with Tensorflow lite). Our neural network uses the TFIDF vectorization technique from scikit-learn to make the text neural network readable). We mainly used Reddit's own CSS styling to make our website similar to theirs. We used Google collab and a notebook on the platform to train and create our models(Vectorizer and neural network). We used the following Kaggle dataset for training: https://www.kaggle.com/clmentbisaillon/fake-and-real-news-dataset. We also made our own logo with https://logomakr.com/.

Challenges we ran into

Deployment was a huge challenge and in the end we could not truly deploy the web-app on a cloud platform. This was due to several reasons and one of them being the complex dependencies and structure we had. Furthermore, our lack of concrete experience in this regard also held us back here. Finally, the icons on the front-page also had to be converted by us into SVG images as Reddit dynamically creates their own(this took a lot of time).

Accomplishments that we're proud of

• The overall polish of the final product.
• The use of production level machine learning techniques and tools.

What we learned

How to use ML constructs in a production environment(and in general). We had some experience with working with them in a more academic to hobby setting but during the hackathon we learn how using tools like Tensorflow lite and Pickle. We also further improved our web development skills by creating the front and backends of this website.

What's next for Aletheia

We would like to add:

• More pattern-based filtering.
• Sorting of posts similar what is done on Reddit.
• User accounts to store preferences.

What the name means

• Aletheia is the greek word for truth and the Greek deity of truth.

Built With

• flask
• html/css
• praw/reddit
• scikit-learn
• tensorflow/keras
• textblob