Aspect-based opinion mining on Yelp reviews
OpenEdge ABL Python Other
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.
app updated app Jul 30, 2014
classes Updating documentation Aug 9, 2014
modeling images? Jul 31, 2014
presentation removed test Jul 31, 2014
.gitignore insignificant tweaks Jul 29, 2014
Beckett's Table.json Temporary sample json summary object Jul 28, 2014 Tweak link May 23, 2017 updated main Jul 31, 2014
yumm_screenshot.png new screenshot for readme Jul 31, 2014

Yelp Summarization Miner (YUMM)

Yelp Summarization Miner (or YUMM) is an automated system that uses machine learning and natural language processing to generate a digestable, human-understandable, and browsable summary of the opinions expressed in a corpus of Yelp reviews about a particular restaurant. YUMM's summary aims to provide the user with an at-a-glance understanding of a restaurant's features (or aspects) as well as reviewers' attitudes towards these features.

You can check out a demo of YUMM online here. If that's not your style, here's a screenshot of YUMM in action (for a restaurant called Zinburger):

YUMM generates a summary of this form in a completely automated fashion from the raw text (and metadata) of Yelp reviews about a restaurant. The items in blue at the top are the "aspects" that YUMM has extracted from the reviews ("service", "food", "sweet potato frieds" etc.). These are the salient features of the restaurant that reviewers often comment on.

When one of the aspects is selected, YUMM displays a summary of reviewers' attitudes towards the aspect. The Positive/Negative bar (see screenshot) displays the overall balance of sentiment that YUMM uncovered in the data with respect to the chosen aspect. YUMM also provides supporting evidence in the form of snippets from reviews.

Code Overview

If you are interested in exploring YUMM's code, the main directories of interest are ./classes, which contains the code for YUMM's primary summary-generation pipeline, and ./modeling, which contains the code for training/optimizing the machine learning models that currently power YUMM.


The problem of automatic review summarization has been addressed in academic literature. See especially: