Queer Recommenders
The skills the authors demonstrated here can be learned through taking Data Science with Machine Learning bootcamp with NYC Data Science Academy.
The Gay Interest:
As a gay man, I am personally interested in the mindset and goals of the gay community. Β And I want to bring awareness to and understanding of this often-discriminated minority group.Β A better understanding of those who are different, will help bridge the gaps in our differences. Therefore, we must βseek first to understand and then to be understood.β (Stephen Covey)
There are also many non-altruistic goals in this analysis. Just being honest here. Gay men often have a large disposable incomes, they often donβt have children.Β Knowledge of the gay manβs interests can be beneficial for companies looking to tap into new revenue sources.
But wait, there's more!Β The data science nerd in me immediately noticed Queertyβs article recommendation system and I wanted to explore how it worked. This could ultimately help the website increase article readership and involvement. Β Double whammy!
This motivated me to scrape data from the popular gay website www.queerty.com, to begin the process of understanding. Get ready to see the unique merger of gays and data science.
Gay-thering Data from Queerty.com:Β
Queerty.com is a news website free of an agenda (except the gay one). Below you can see the basic layout of the website. There are 4 different categories: "Life", "Entertainment", "Politics" and "Goods" (goods referring to hot guys) highlighted in the top left red box. Using Selenium with python I would crawl through everyΒ article in the webpage.
A typical structure of a Queerty article isΒ shown below. I collected the following information about every article on the Queerty
- Category: String (βlifeβ, βpoliticsβ, βentertainmentβ, βgoodsβ)
- Author: String
- Title: String
- Number of Comments: Float
- Tags: String separated with β|β
- Article URL: String
- Recommended URLs: String separated with β|β
- Time: String -> converted to datetime
- Number of Tags: Float
- Title Length: Float
In total, I scraped 12,079 articles with 6 features directly from the Queery website and took several days to collect. To quote a colleague, "Selenium is sooooo slow!"
Gay Men are Talking Life and Politics:
An exploratory data analysis can often provide insightful information.
After a logarithmic transform, we see a fairly normal distribution in the number of comments as shown on the right. Β Of course we want to explore how the comments are distributed among the different categories, shown below.
The following bar and box plot clearly indicate the main interests for the gay man are politics and life issues. Way to stay informed, ladies! Personally, this made me very happy. Now we need a section on gay men in science and math....
Queerty Diagnostics:Β
This section provides interesting information about the Queerty website
I also examined the cumulative sum on the number of comments since 2006. The interesting aspect here is the slope of the curve to the left.
One can see a dramatic increase in the slope of the curve since late 2016. Β If I had to guess, gay men are venting about the Trump election.
Next we will attempt to understand the recommendation system used by Queerty. Therefore, I Β linked the main article urls to all the recommended urls.Β First of all, Β 49.9% of the recommended articles are Queerty articles and the rest belong to an array of other
Surprisingly, the tags feature are not used to recommend articles to the reader. As you can see on the right, the majority of articles have 0, nada, zilche common tags. It left me thinking, "That's a Queer Recommender."
How does the recommender pick articles for the user? The next obvious link had to be the title of the articles. And bingo! After removing common stop words, you can seeΒ in purple, there are many common words in the titles of the main article and the recommended articles.
Moving Forward:Β
The focus of this project was to develop our skills in scraping a webpage and to perform an exploratory data analysis. There is a lot more information I would have like to have scraped, such as the comments themselves. I think it would be informative to understand the comments using a sentiment analysis. Also, I would like to explore just the top, visible, 6 articles of the recommender system. It would be interesting to see how changing the ratio of Queerty and non-Queerty articles effects the readership and the bottom line. Queerty, hit me up if you want to explore together!