FLAVA is based on the theory that the words people use when describing their experiences often reveal a much more accurate depiction of true thoughts, feelings, and emotions compared to merely a number rating.
Many review sites offer reviewers a chance to assign a 1 to 5 rating to a local business, often accompanied with a text description with more details of their experience. Simple number ratings can often be unreliable due to several factors: they don't capture the true sentiment of the reviewer, and reviewers may actually change their mind on the final number they assign after having the opportunity to dump out all of their thoughts. However, if we can determine a rating based on text alone (taking the ability to choose an arbitrary number rating away from reviewers), we suspect that we will get a more accurate review overall.
What it does
FLAVA pulls in reviews from leading sites such as Yelp and Google Maps. Review text is then run through the expert.ai Natural Language API to determine the sentiment score for each piece of text. The average sentiment value of the reviews is then used to give the final FLAVA rating, which will categorize local businesses within one of the four categories using a weighted algorithm: LOVED, LIKED, MEH, or HATED. This gives end users a quick and easy way to determine whether or not they may enjoy being a customer at any local business.
How we built it
FLAVA was written in Python and takes advantage of the Python library for the expert.ai Natural Language API. I also developed a frontend using HTML/CSS/JS with a bit of PHP for final score processing.
Challenges we ran into
This was my first time writing a Python web app, so I learned a lot about Django during this process. I also created a standalone Python script that is currently on GitHub for testing purposes. Currently using Heroku for Python hosting.
Accomplishments that we're proud of
FLAVA has the potential of becoming integrated with other services and platforms. People who typically seek out detailed reviews might be pleasantly surprised. Checking FLAVA is quick and easy if you don't have time to sift through multiple reviews.
What we learned
I learned more about more about sentiment analysis and how it works. I first briefed myself on the Wikipedia article here, and then found it a very interesting experience to actually apply some of these concepts and see the results: https://en.wikipedia.org/wiki/Sentiment_analysis
What's next for FLAVA
I would like to incorporate more review sources and expand upon the front-end interface for FLAVA to make it more engaging and informative. I also want to improve the search function and incorporate predictive text to make it more user-friendly.