I built Yourbnb, an NLP-powered web-app that mines Airbnb reviews to find listings that align with user preferences.
Web address: www.yourbnb.xyz Link to google slides presentation: https://bit.ly/2qcMbVI
Approach:
(1) Train a Latent Dirichlet Allocation (LDA) topic model on Airbnb reviews from listings in NYC.
(2) For each listing, get a vector represeting the ratio with which each topic was discussed in the listings' reviews.
Save the Airbnb listing data and LDA model as pickles in the directory: 'airbnb_insight/Flask/MVP/'
(3) Validate the model by checking that the topics discussed within listings are more similar than those between listings.
(4) Deploy a Flask web app (hosted on AWS EC2) that takes a user's input, applies the same LDA model to it, and ranks listings by their similarity to the topics discussed by the user.
Code for steps (1) - (3) are included in the jupyter notebook 'airbnb_insight/Topic Analysis.ipynb'
Code for the Flask app is in: 'airbnb_insight/Flask/MVP/'
• Python code for the app is: 'airbnb_insight/Flask/MVP/views.py'
• Helper functions for this app located in: 'airbnb_insight/Flask/MVP/a_Model.py'
• Input and Output pages: 'airbnb_insight/Flask/MVP/templates/input.html' and 'airbnb_insight/Flask/MVP/templates/output.html'
Airbnb listing and review data were downloaded from http://insideairbnb.com/get-the-data.html
Reviews: reviews.csv.gz under the 'New York City, New York, United States' heading
Listings: listings.csv.gz under the 'New York City, New York, United States' heading
This app has not yet been optimized for running on someone else's machine