Coursera Web Scraping Catalog
Coursera is the world's biggest MOOC (Massive Online Open Course) provider with 37 million registered students spread over 4,000 courses. Founded by computer science professors, Andrew Ng and Daphne Koller from Stanford University, Coursera offers a chance for anyone with an internet connection to learn at their own pace.
I wanted to explore and better understand the online course marketplace through the simple question: what makes a student choose a particular course?
To dive deeper, I asked three specific questions of my data:
- Do users choose courses based on the university / company / partner that provides the content?
- Is name recognition and prestige a primary consideration in enrolling for a course?
- Do ratings matter? Are the highest rated courses the most enrolled in?
- Does the career growth of others drive enrollment in a particular course?
- Coursera displays a metric for career benefit based on past surveys for each course.
Web Scraping Technique
To get the data on each course, I created 2 spiders in Scrapy; one to go through the course catalog and one to explore the course sitemap (an afterthought when I found the course catalog limited the number of courses displayed to only 1000).
For each course I looked at the following primary attributes for my analysis:
- Name of Course
- Number of Reviews / Ratings
- Career Benefit Percentages
- Total Enrollment (response variable)
By grabbing data from courses by section, I had to do a substantial amount of data cleaning to get the information into a workable form. One of my main challenges was handling multiple different features combined into a single column (ex. "Difficulty Level," "Language," "Approx. hours to complete" all in the same cell). I addressed this by using regular expressions to filter through and categorize specific results into new columns and performing a gather and spread operation to capture certain categorical variables.
Ratings: Do ratings matter for enrollment? At a very preliminary level, it does not appear ratings have much influence on course enrollment. All courses seem to have had an average rating of 3.0 or higher and the spread of enrollment did not seem predisposed toward the highest ratings. Without accounting for number of courses in each category however, this could simply be a result of most courses being within a certain spread of reviews.
Partner: Does the brand name of the course creator matter? It does appear certain universities (some unexpected ones) have a large segment of the overall user base of Coursera. Besides some Ivy League schools like Stanford University (popular, in part, due to their length of time one the site and number of offerings), it is interesting to note that universities from outside the United States break into the top 10, suggesting the website use is not limited to US students.
Career Benefit: Looking at courses that reported career benefits (3 kinds) and comparing them all with enrollment seemed to show no substantial increase in course signup over differing proportions of career benefit. This is interesting because it appears Coursera uses this as a method to entice enrollment. The effect it has, I would assume, is that it encourages enrollment without comparison of the actual percentage number. Put another way, the idea of a career benefit is more useful than the percentage of people who claimed it was.
With more time, I would like to go deeper into my analysis of these multiple attributes (ex. Approx length of course in hours) and better account for multiple factors that would affect my current data to get more solidly built conclusions. Ultimately, it may be an interesting practice project to train a machine learning model to predict a course's enrollment based on existing data. Web scraping other MOOC providers would also give me a bigger picture overview of the online course landscape in general.