I’ve rebooted the Yelp Summarization Miner (YUMM) project. YUMM uses natural language processing and machine learning to automatically generate an aspect-based sentiment summary from the raw text of Yelp reviews.

You cand find a demo of the project live here (takes a minute to load the first time), or find more details on github. Here’s a screenshot of YUMM in action for a restaurant called Pizzeria Bianco:


As you can see, YUMM attempts to extract relevant “aspects” (“margherita pizza”, “fennel sausage” etc.) from the raw text of Yelp reviews about the restaurant. It then aggregates review sentiment expressed towards that aspect (e.g. 12 positive reviews, 5 negative reviews), and provides supporting evidence (e.g. “Best margherita pizza I’ve ever had”).

The full code for YUMM is available in the github. YUMM is heavily inspired by Blair-Goldensohn et al.’s “Building a Sentiment Summarizer for Local Service Reviews” (2008).