You are what you read - read choices reveal interesting latent features


You Are What You Read

Can our reading choices reveal something about ourselves?
 



“Tell me what you eat, and I’ll tell you what you are” Brillant-Savarin”:  Iron Chef


The renowned scientist Jean Savarin once said, “Tell me what you eat, and I’ll tell you what you are”. We can extend this notion outside of food, to anything we consume. Whether it’s food, music, or what we read. Our choices reflect our preferences. Companies like Amazon, Netflix and Spotify use preferences to make predictions. These predictions are used to market to you what you may like in the future. The goal of making these predictions is to help provide better customer experience. For example, to make a better song recommendation on Spotify. Or to sell an additional product on Amazon. Then, can we say the same for our reading behavior. Namely, can we know a person better based on what they read?
Imagine this classic analogy, you are in a supermarket. The aisles are content topics, and your basket is what you choose to read. From news, sports to fashion, you browse this supermarket. Picking the stories and articles that you wish to read. and put it in your basket. Your basket represents your personal combination of preferences. By combining these preferences, allows for grouping of preferences. These grouped preference, can used to build audience affinities. These audience affinities happen naturally, as people tend to seek groups that share their interests. And uncovering these audience affinities is a common practice in data science.
For example, Spotify or any other music service, has a record of every song you have listened to. Spotify also keeps a record of everyone else and their musical history as well.  Music services like Spotify know that a happy customer is more likely to stay a customer.  To make their audiences happy, and keep their product ‘sticky.  ‘Spotify has to keep their product relevant to their customer. To be relevant, they aim to expand the user’s musical taste. To do so, they first try to find users with similar music tastes to build a music segment.  With this segment, they find other music that was listed to by the group. They use this selection of music to recommend to the group. These recommendations expand as new music exposure to their users. Spotify use recommendation engines to find similar music or content you may like.

            Here is an example of a recommendation system.  Say I like sports articles, and Jack and Diane also like sports articles.  Jack and Diane also like gossip news. If Jack and Diane are similar to me, then I also may like gossip news. Of course, it depends on how similar these groups are. If we were able to hone into deeper detail, we can unlock even more data potential. Instead of sports articles, I was reading hockey news for the Pittsburgh Penguins. Then, I would have to find people who were also into the Pittsburgh Penguins. Once I found those users, see what other topics they read.  This method unlocks latent information from similar audiences. The topics that Jack and Diane like outside of our shared love of the Pittsburgh Penguins, I also may like. This methodology is known neighborhood approach. A technique that is commonly used in recommendation systems.


A common recommendation classification algorithm.[i]

While we won’t go into the math behind this. We can assume that these method helps managing a very large data problem. To apply it to our reading scenario. We simply look all the traits of articles a person reads, apply our recommender system, and out comes the results. This application can be used to fostering a deeper connection with a audience.
Publishers can utilize recommendation engines, to discover latent traits of their readers.  By doing so, they have the potential to unlock a wealth of data opportunities.
One use would be to automatically critique and serve content based on my interests.  Another use may be to display a relevant sponsored message. Nevertheless, publishers can use this to make their product more 'sticky'. Just like the shows we watch, or music we listen to. Optimizing articles based on past preferences can elicit a better user experience. By doing that, you are indeed, what you read. 





[i] Deep Neural Networks for YouTube Recommendations: https://static.googleusercontent.com/media/research.google.com/ru//pubs/archive/45530.pdf

Comments