Inspiration and what it does
Throughout my childhood, one of my favorite ways to spend quality time with my mother was by editing my writing. Whenever I had an essay due I would curl up next to her on the couch and learn how to improve my articulation tenfold.
In today’s media-saturated world, article authors and suppliers need every edge they can get. Unfortunately, not all purveyors of the written word are blessed with a mother as wonderful as mine. That is why we created an AI that can analyze real estate articles based on text.
Our AI can take in the text of an article and predict both a rough estimate of the number of hits it will get (based on percentile) and the tags that you should add to the article.
How we built it
Our first step was to run a number of analyses on the data using a variety of neural networks and techniques. For example, we tried running elasticnet cross validation regressions using quantitative data extracted from the text. Unfortunately, the root mean squared error was ridiculously large. We eventually settled on using analyses based solely on article text for our project because: They were able to reach 80%-90% accuracy. It is easy for a user to input a single block of text, rather than fill out a series of values.
We found that the most accurate approach was to invoke a multi-label classification technique cleaning the data with NLTK and using a scikit-learn SCV_Pipieline to analyze the data.
Challenges we ran into and accomplishments we're proud of
Our biggest challenge was that (being beginners) none of us have experience with any front end software or techniques.
We were very proud of the way that our team learned HTML, CSS and Flask from scratch in less than 24 hours!
We were also impressed with the way (despite our inexperience) that we were able to successfully implement a multi-label classification neural network model. That is no small feat.
What's next for Apex Article Analytics
Going forward, Apex Article Analytics would love to spend more time using techniques like blackout feature selection to help our website not only provide general feedback and article analysis but to offer specific advice on how a given article could improve.