Multi-Domain Yelp Business Reviews

A multi-domain sentiment classification dataset extended from the Business and Reviews datasets supplied by the 2015 Yelp Dataset Challenge.

Prerequisites

The parsing functionality requires both datasets to be downloaded and supplied to the parsing script.

The following Python packages must be installed:

• Pandas
• Numpy
• Datasets (HuggingFace)

Usage Instructions

Run parse.py with the following, mandatory arguments:

• -rf or --reviews-file: Filepath to the Yelp Reviews JSON file (str)
• -bf or --business-file: Filepath to the Yelp Business JSON file (str)
• -out or --output-dir: Filepath to the directory where the datasets will be saved.

Domain Metadata

Domain	Positive-class Samples	Negative-class Samples	Total Samples
Restaurants	1,673,359	971,671	2,645,030
Food	698,971	420,512	1,119,483
Nightlife	579,360	306,257	885,617
Event Planning & Services	199,582	146,431	346,013
Shopping	120,920	142,777	263,697
Hotels & Travel	98,421	108,977	207,398
Arts & Entertainment	120,967	66,981	187,948
Beauty & Spas	56,199	92,444	148,643
Automotive	28,589	91,244	119,833
Home Services	22,972	91,698	114,670
Active Life	55,747	39,946	95,693
Local Services	26,406	67,872	94,278
Health & Medical	22,861	63,712	86,573
Local Flavor	50,576	19,919	70,495
Real Estate	7,612	27,346	34,958
Pets	11,188	21,926	33,114
Professional Services	6,374	20,134	26,508
Public Services & Government	10,922	7,469	18,391
Education	7,398	6,445	13,843
Financial Services	2,280	11,276	13,556
Religious Organizations	1,625	657	2,282
Mass Media	926	1,320	2,246