| output | puppeteer | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
pdf_document |
|
What is it?
forecasting restaurant success based on menu items and yelp reviews
Class: CS 410: Text Information Systems Group Project Authors:
- James Banasiak [email protected]
- Mark Hornback [email protected]
Date: 10/12/2019
This code and all materials (except data sets) will be stored at the following [private] repository location. Request permission to the repo by emailing the above.
git clone https://lab.textdata.org/jamesmb3/foodcasting
Google Slides public Presentation.youtube Presentation.mp4 requires @illinos.edu identification
data.zip requires @illinos.edu identification.
unzip to data directory - then you can import it (below) under Rebuilding local data
cd foodcasting
unzip data.zip
To reproduce this report in its entirety, the following tools must be installed.
- Atom - A hackable text editor for the 21st Century download
- used Markdown Preview Enhanced package for markdown and prince for pdf conversion.
- used mermaid for diagrams
Restaurant menus and yelp reviews have been a target for testing various computational tasks because the data is publicly available. We plan to utilize menu data and yelp reviews data to attempt to predict the likelihood that a restaurant will have positive reviews based on the menu provided.
We plan to achieve this by creating a language model for the menus and separately performing sentiment analysis for reviews. We will take into account document length and term frequency.
Professor of Computer Science at Stanford University Dan Jurafsky has authored several papers on the subject showing how word choice distinguishes price and how low one-star ratings were related to traumatic experiences and low sentiment.
Narrative Framing of Consumer Sentiment in Online Restaurant Reviews
We will attempt to further this research by utilizing data sets and providing a reverse mechanism to estimate sentiment based on the menu contents.
This tool is designed for restaurants to help distinguish which menus and menu items may be more successful than other menus and items. The individual descriptions for menu items and word choice can play a key role in determining price and positive sentiment of reviews, which translates into success of the restaurant. This will also potentially help restaurant suppliers and/or investors focus their energy on restaurants that will be more successful. If time permits, this can also be extended into item ordering and layout as well.
As of the time of our research, it appears that no such tool does exist.
- Logistic regression with the SGD optimizer
- Logistic regression with the LBFGS optimizer
- Support Vector Machines
- Decision Trees
- Gradient Boosted Trees
- Random Forests
- Naive Bayes
- NTLK
- Valence Aware Dictionary and sEntiment Reasoner (Vader)
We will demonstrate a user calling our application and passing it a menu file. The output will be a yelp star value that we are most confident with.
Given a menu as input, provide a yelp star prediction based on a model created from our corpus of menus and reviews.
The program can be run via commandline or API. We will collect menus and corresponding yelp reviews and split them into training sets to tune our model.
Given a menu as input, provide a yelp star prediction.
The description of the elements contained within the data set is:
For the business listings -
-
slug- a unique portion of the url, typically dashed business name for http://{host}.com/path/{slug} -
categories- the type of foods the restaurant serves, eg Pizza, Chinese, Mexican -
distance- the distance from the centroid of the zip code -
name- the name of the restaurant -
price_level- a yelp categorical price level $= under$10. $ $=11-30. $$$=31-60. $$$$= over 61 USD -
rating- predictor the yelp rating associated with a restaurnat -
review_count- the number of reviews that were taken to establish the rating -
url- the url from yelp -
lat-the latitude of the location of the restaurant -
lng- the longitude of the location of the restaurant -
Sp1- a spacer (not used) -
type- the type - allnaturalwords -
homeurl- the path portion of the url -
resource_id1- a resource id used in yelp specific api -
resource_id2- a resource id used in yelp specific api -
lat2- the latitude again resulting from a join within scraper -
lng2- the longitude again resulting from a join within scraper
For the menuitems-
slug- a unique portion of the url, typically dashed business name for http://{host}.com/path/{slug}title- a menuitem title egChicken Caesar Saladdescription- the longer description of the title egGrilled chicken, romaine, Parmesan, tomatoes and Caesar dressingprice- the price of the menuitem eg7.99
You can download a smaller subset of the chicago data that we used to initially test. The entire data.zip file is provided above for the full data set that was used in the project. chicago-menu.csv chicago-summary.csv
Why sqlite3 database?
Here is a good article that explains sqlite vs pandas and performance gains.
In addition joining is much faster and we can do more complex things within sqlite3.
# clone repo
git clone https://lab.textdata.org/jamesmb3/foodcasting
# change dir
cd foodcasting
# setup sql lite to setup database
sqlite3 sql-lite-cache/foodcasting.dbRun the create-db-yelp.sql script from within sqlite3
sqlite> .read create-db-yelp.sql
- Clean tables
(base) > $ sqlite3 sql-lite-cache foodcasting.db
SQLite version 3.30.0 2019-10-04 15:03:17sqlite> DROP TABLE summary;
sqlite> DROP TABLE menu;
sqlite> .read create-db-yelp.sql
sqlite> .quit- Import compressed summary from shell
gunzip -c ./data/summary.csv.gz | sqlite3 -csv -separator ',' sql-lite-cache/foodcasting.db '.import /dev/stdin summary'- Import compressed menu from shell
gunzip -c ./data/menu.csv.gz | sqlite3 -csv -separator ',' sql-lite-cache/foodcasting.db '.import /dev/stdin menu'- Verify counts
(base) > $ sqlite3 sql-lite-cache/foodcasting.db
SQLite version 3.30.0 2019-10-04 15:03:17
sqlite> select count(*) from summary;
65480
sqlite> select count(*) from menu;
2244981
sqlite> .quit- Optional - import the census data used to determine top 1000 zip codes.
gunzip -c scrapy_yelp/data/population_by_zip_2010.csv.gz | sqlite3 -csv -separator ',' sql-lite-cache/foodcasting.db '.import /dev/stdin usa_2010_population_by_zip'brew install tesseract
pip install tesserocr
pip install Pillow
Use the notebook ocr/menu-ocr and run each step.
Using a Sagemaker deployed model, a lambda was created to invoke the endpoint.
The lambda code can be viewed at
/aws/lambda_function.py
A public facing API endpoint was configured to access the lambda, allowing the public to access our model with a simple POST request, with the data in the request body. (Strip or delimit commas from the menu text). The body contains a data object of “menu text, category, restaurant name, price, review count”
https://q3di7y5jsj.execute-api.us-east-1.amazonaws.com/yelp-test/yelp-menu Body:
{"data":"Wine $60 steak $65 pasta,Steak,Genes steakhouse,$$$$$,1000"}
In the presentation we use different requests to get the values of rating of the menu being.
curl --header "Content-Type: application/json" \
--request POST \
--data '{"data":"Whopper $5 Fries $2 Shake $4,Fast Food Burgers,Burger King,1,100"}' \
https://q3di7y5jsj.execute-api.us-east-1.amazonaws.com/yelp-test/yelp-menu
3.9464221000671387%
curl --header "Content-Type: application/json" \
--request POST \
--data '{"data":"Whopper $5 Fries $2 Shake $4,Pizza,Pizza,5,1000"}' \
https://q3di7y5jsj.execute-api.us-east-1.amazonaws.com/yelp-test/yelp-menu4.17765474319458%
This report acknowledges Professor ChengXiang Zhai and the wonderful TA's for their advice and expertise in within Intelligent Information Systems and CS410 Text Information Systems.