Sentiment Analysis and Data Exploration of Yelp Reviews
The skills the author demoed here can be learned through taking Data Science with Machine Learning bootcamp with NYC Data Science Academy.
Background:
Restaurants are constantly getting feedback. Yelp is one channel, but many restaurants document verbal feedback, gather feedback from questionnaires, or even from other review sites. When faced with a huge amount of text, restaurants need a way to objectively interpret their reviews. It is very difficult for business owners to go through list of reviews and distill relevant data and information from it.
Also it does not provide advanced analytics to business owners to grow their business and improve their services. My solution will focus on providing advance analytics from Yelp data to help existing business owners, future business owners and users.
The objective is to design a system that uses existing Yelp data to provide insightful analytics and help both existing and future business owners make important decisions about launching or expanding their business. It provides opportunity to business owners to improve their services and users to choose the best business from the available options. By building a way to relate text-based feedback to a star rating, we can help these restaurants understand how they would rate on Yelp.
This post is about using the web scraped online reviews from Yelp to have a better understanding of the industry. These review analysis can help the customers what to expect from a particular restaurant and also business can look into the different opinions and work on the aspects that could possibly lead to a low rating.
Approach:
I divide this project into 2 segments based on different objectives.
- One is to assist future entrepreneurs in defining probability of success prior to launching a new venture in a specific city. They can also determine which attributes (Location, Pricing, Cuisine type etc.) play a major role in attracting more customers.
- This should enable existing business owner use analytics to to improve their services and to make better decisions regarding business expansion in new cities by performing sentiment analysis on the poorly rated reviews.
Project 1: Data Analysis and Visualization of Pricing and Rating Info
Data Collection
- The project is coded in both Python and R.
- There are close to 50 pages, each having links to 30 restaurants, within which the individual reviews are listed.
I started with recognizing the pattern of 50 pages and implemented the same in my algorithm to set those as the starting URLs in a list comprehension. Most of the time, it wouldn’t scrape beyond the second page. The whole process of web scraping (Yelp website) was quite a challenge but rewarding.
Cleaning and Repair
While scraping the Pricing and Rating information, I ran into scenarios where the Pricing was not listed for a few restaurants, which resulted in some missing values. These missing values are imputed by using the high frequency value for pricing.
As the data got bigger, the categories also increased for each variable. In order to visualize the values, we had to restrict the data to limit to top 15 places in the city with the most restaurants.
Exploratory Data Analysis (EDA)
Most of the attributes were obtained in an easy manner. However, Pricing posed some challenges, as some restaurants were missing this information. Also the class information for Location was not consistent across the restaurants, causing us to handle these as exceptions.
try:
Price = review.find('span', attrs={'class': 'business-attribute price-range'}).text
except:
Price = ""
try:
location = review.find('span', attrs={'class': 'neighborhood-str-list'}).text.strip()
except:
location = ""
I got started by looking at the places in NYC, which tend to be expensive. Below is the barplot of these attributes. As we see, West Village and Greenwich Village has the high volume of expensive restaurants.
Are these expensive restaurants really that good?? Hmm…may be not…
Here is a plot of pricing and Rating of these restaurants. Surprisingly, restaurants which are more reasonably price do seem to be having the highest rating. Does that mean that more and more customers visit the places that are more affordable, causing them to have good rating? That is one possible cause.
Now let’s see where are these highly rated restaurant, which are also affordable are located.
The Lower East Side is the place to be. This place has the top most rating (a rating of 5.0) followed by DUMBO. The Majority of the restaurants in the financial district seem to have a very poor rating.
These visualizations will enable existing and future business owners to decide to what extent they need to consider location in their plans. This would help them identify good locations in which tor open their first restaurant or branches of their existing business .
Project 2: Sentiment Analysis of Review Content
Overview
Sentiment analysis is the computational task of automatically determining what feelings a writer is expressing in text. Sentiment is often framed as a binary distinction (positive vs. negative), but it can also be a more fine-grained, like identifying the specific emotion an author is expressing (like fear, joy or anger).In our analysis, I applied both forms of sentiment analysis, i.e., the traditional and the NRC.
There are many ways to do sentiment analysis, though most approaches use the same general idea:
- Create or find a list of words associated with strongly positive or negative sentiment.
- Count the number of positive and negative words in the text.
- Analyze the ratio of positive to negative words. Many positive words and few negative words indicates positive sentiment, while many negative words and few positive words indicates negative sentiment.
For example, "sick" is a word that can have positive or negative sentiment depending on what it's used to refer to. If you're discussing a pet store that sells a lot of sick animals, the sentiment is probably negative. On the other hand, if you're talking about a skateboarding instructor who taught you how to do a lot of sick flips, the sentiment is probably very positive.
Data Collection
Since we are only interested in 4 highly reviewed restaurant, we treated these separately. The 4 restaurants are Eataly, Morimoto, Tao Uptown and Ilili. Scraping them individually also posed some challenges due to the limits beyond certain pages. The reviews are sorted by low ratings, and these pages were scraped to analyze the sentiment behind these.
The reviews from these 4 restaurants are then imported into R and combined into a single file.
Data/Text Cleaning:
The review contents is then cleaned to remove any punctuations, junk character, white spaces and new lines. The stop words like ‘I’,’ me’, ‘we’, etc. were also removed as they wouldn’t add to our sentiment analysis.
I generated a word cloud of the 100 most frequently appearing words across all restaurant review text,to get an idea of the vocabulary people tend to use in their reviews. I noticed a lot of the bigger words seemed neutral, but we do see words like “time, dinner, disappointing, wait”etc. It was also interesting to see words such as “manager,, “crowded,” and “price,” appear large in this cloud, as it gave us an initial indication of things that might matter a lot to reviewers.
Below is the bar plot of the sentiment analysis:
Conclusion:
Future entrepreneurs always to consider where and what kind of business one should open to gain maximum profit. Using the above proposed solution, business owners will be able to determine what kind of business is more profitable, sustainable and attract more users in a particular city or area. In addition, they will also be able to determine what type of restaurants (e.g., Mexican, Chinese, etc.), price range and location are more favorable to succeed in particular city.
Link for the GitHub Code:
https://github.com/KiranmayiR/Yelp_Web_Scraping
About Author
