Perspectives

Perspectives Logo.
An increasing political divide, particularly in online media, serves as the inspiration for this project.
The two robust AI models integrated into the Perspectives project.
Perspectives generated recommendations for left-leaning article.
Perspectives generated recommendations for neutral article.
Perspectives generated recommendations for right-leaning article.

Inspiration

Following the presidential elections in a very politically polarized era, we wanted to address some of the difficulties that we faced in finding a good balance of political news sources.

What it does

Using deep learning, this extension helps you understand the political bias (left, right, neutral) of news articles you are reading. It then encourages you to explore articles of opposing perspectives, which were identified using natural language processing.

How we built it

The Chrome extension extracts the URL and contents of the current news article and sends a get request to a Flask server running on Python. The server analyzes this request and forwards it to a bidirectional LSTM neural network that returns the bias score on a spectrum. The backend script also analyzes the article using natural language processing and scrapes articles with different opinions on the same story.

Challenges we ran into

Text is a messy data format to deal with. As a result, a large portion of our time was spent parsing through inconsistencies in the data and formatting everything consistently. Also, it took several iterations of hyperparameter tuning in order for the model to achieve the level of accuracy that we reported.

Accomplishments that we're proud of

This was our first time working together and we are happy that we managed to pack a large number of sophisticated functionalities into a project with large real-world applications.

What we learned

Working with language models can often be quite challenging due to the erratic nature in the text data and their meanings. We learned to focus on streamlined processing techniques to minimize the amount of unwanted variability in our results.

What's next for Perspectives

Continued hyperparameter experimentation will lead to higher classification accuracy of articles. Integration of cloud-based platforms will allow for larger requests. Eventually, Perspectives could be introduced to the Google Chrome Store for public use.