Skip to content
This repository was archived by the owner on Oct 8, 2021. It is now read-only.

ReviewTruncation

Curtis edited this page Mar 9, 2020 · 1 revision

Situation

I wanted to incorporate tensorflow.js into my project and use sentiment analysis on Yelp reviews to assist users using ChowMapper.

Problem

Yelp's Fusion API returns only 3 reviews, and they are truncated to few enough characters to not properly train any sentiment analysis model. These reviews throw off any model.

Solution

Google Places Example data You can leverage Google's Places API and retrieve up to 5 full reviews of a place. You can get the Google Place most similar to the Yelp business by converting the Yelp-provided phone number to the E164 format and giving it the Places Search endpoint. Using the first returned result (it's usually one), the business is directly looked up, asking ONLY for reviews to save GCP credits

Solution problems

Yelp shows some parks as a business, which Google Places does not recognize.

Clone this wiki locally