Motivation

Though it may not be representative from a statistics point of view, people often refer to comments/reviews from website, forums, and social media platforms as reference when making decisions for a variety of use cases. For example, people may read reviews on yelp.com when picking restaurants, or amazone.com when making purchases of consumer products. While most of these website provide quality overviews in the form of out-of-x-star ratings, I've yet to see a platform that provides collective overview of customer sentiments directly from those reviews that they left. Real sentiment ratings, though still suffering from statistical biases, may be more useful than star ratings when basing decisions upon because it provides more information overall and the out-of-x-star rating system is less nuanced in capturing sentiment.

This project, in its ultimate form, will automatically analyze all human comments/reviews to provide customer sentiment rating / graphical representation given a link to any website. Please refer to the "version History" section to see what has already been implemented.

​

​

​

​

Data

Current data are collective for a for particular business listed on yelp.com.

(Please open the code file to see details).

​

​

​

Algorithm

Current version leverages the BERT model from huggingface to conduct sentiment analysis.

​

​

​

​

Version History

 

• WebR0.0: Implements minimum working pipeline to scrape all reviews across all pages from a particular website for initial testing of rating outcome and data visualization

​

Future Versions

​

• WebR0.1: Introduces capability of recognize (distinguish) human language (reviews) out of all texts on a particular website

• WebR0.2: Realizes all functionalities from previous versions to any (appropriately structured) websites (significant scraper bot challenge)

• WebR1.0: Deploys on cloud and connect to web domain for general users to use