Inspiration
What it does
Uses a keyword extractor and ML to generate projected tweets and favs based off of keywords and scaled to Twitter user's followers.
How I built it
The first step was employing Twitter's API to obtain large sample sets (between 4k-12k) of tweets from Twitter's live stream. The tweets were then run through a keyword extractor script which we adapted from MonkeyLearn's API in order to identity the keywords and relevant topics from the rest of the text. This information, coupled with the number of Favourites, Retweets, and user Followers, was then fed into Microsoft Azure's Machine Learning Studio as a csv file. The number of either Favourites or Retweets served as the dependent variable (the independent variables being the number of followers held by our given user and the specific keywords employed in the tweet). From there, we implemented a 70/30 data split, testing the 30% untouched data against the predictions generated from the 70% of data fed into Azure's Poisson Regression algorithm.
Challenges I ran into
Despite the user-friendly interface and streamlined tutorials, Azure's limitations as a beta platform induced a fair number of unexpected setbacks. The range of different applications implemented in our model produced a large degree of cross-compatibility issues, requiring the majority of scripts to be rewritten, adapted, and tested for legitimate compatibility. (This took a very, very, very long time).
Accomplishments that I'm proud of
We're quite proud of the way in which we came together as a team, from meeting each other for the very first time at opening ceremonies to presenting our hack at the end of the weekend. Despite the wildly disproportionate ratio of roadblocks faced to hours slept, each new challenge was met with determination. The amount of fast-paced learning (see below) done this weekend was also, in our view, something to be quite proud of.
What I learned
Both members of our team learned a great deal about the key concepts integral to Machine Learning processes, as well navigating Microsoft Azure's ML environment. In regards to complications from requiring each of us to pick up a handful of new skills on the fly as we attempted to circumvent the issues.
What's next for Avivore
There were a good number of features, including and associated keyword generator for improving Tweet visibility, and a sentiment analysis model for analyzing the projected responses predicted by a given Tweet.
Log in or sign up for Devpost to join the conversation.