AirbnbRecommend

Airbnb Recommender System on Boston data using Sentiment Analysis and Collaborative Filtering

Motivation

Goal: build a Recommendation System with publically available Airbnb data from Boston. This will mean being able to recommend Airbnb listings to users based on their preferences.

So, how do we do that? We'll attack this problem in three parts:

1. Determine user preferences - Sentiment Analysis

• Do the necessary text pre-processing steps for each reviewer's comments (including language detection)
• Using NLTK's Vader lexicon, calculate the compound polarity of each comment

for our predicted polarity values, I additionally created this dashboard app with Streamlit and deployed on Heroku

2. Build a recommendation engine - Collaborative Filtering

• Once we have the polarities for each reviewer/listing pair, build a utility matrix with reviewer_id on one axis and listing_id on the other
• Predict all polarities (fill in the Nan values of that matrix) using SVD

3. Generate recommendations!

• We can chose the number of top recommendations we want to give a particular user, sorted according to their predicted polarity

File descriptions

This repository includes:

• data folder, including reviews.csv and listings.csv.
• Airbnb_SentimentAnalysis.ipynb for part 1
• Airbnb_CollabFiltering for parts 2 and 3