BLOG

Recommender Systems are Here to Stay

It’s hard to surf the net these days without a clever recommendation tempting you down a rabbit hole of creditcard-sucking clicks or the YouTube play button. While recommender systems seem relatively new and sophisticated, it has taken a long time for the Internet to catch up to our offline habits of sharing useful information. Humans, for example, were recommending watering holes and berry bushes to fellow tribe members in prehistoric times and well before the dawn of the Internet.

Author and marketing guru Seth Godin talks about the power of word of mouth in what he calls the “pre-TV industrial complex.” [1] Godin explains that prior to TV advertisements, we turned to a trusted craftsman to recommend a certain type of cheese or clothing based on our individual needs. TV ads and mass retail then ushered in a new era of blind consumption and stole attention away from the advice of traditional industry experts. But despite the broad influence of TV commercials to recommend products and services to a general audience, they never replaced the relevance and accuracy of recommendations from friends and family.

Like television, the limitations of the Internet—at least initially—was that it didn’t know or understand our individual needs. Moreover, average users felt disinclined to trust the Internet and especially in its early years. Today, shady corners of the Internet remain but thanks to the sophisticated use of data and advanced algorithms, the Internet typically knows our unique preferences better than our friends and family—and sometimes even ourselves.

In the book, Everybody Lies: Big Data, New Data, and What the Internet Can Tell Us About Who We Really Are, author Seth Stephens-Davidowitz likens Google to a “confession box,” a place for us to tell the web our darkest secrets and unsaid desires.[2]

I recently sifted through my own personal Google settings by navigating to the “Privacy” tab and then to an innocent blue hyperlink called “Google Ads Settings.” This redirected me to the landing page https://adssettings.google.com and a section titled “TOPICS THAT YOU LIKE.”

 

Figure 1: TOPICS THAT YOU LIKE, Source: Google Settings

 

Before me on the screen was a list of 60 pre-populated categories. Like my friends and family, Google knew about my penchant for “Cricket” and “Dance and Electronic Music” and my work-related interests in “Cloud Storage,” “Distributed & Cloud Computing,” and “Computer Hardware.” This was accurate but nothing my friends, colleagues, and family didn’t already know.

Google, though, appeared to know about my recent interest in Digital Currencies—something my parents don’t know. Then I looked even closer but what I saw made me start to doubt the great search engine. Why was “Classical Music” on the list?

I had no memory of ever searching for “classical music” or visiting a website of that genre. And then it hit me. YouTube.

I had listened to tracks from this music genre on YouTube while writing my first book in late 2016. YouTube and Google are joined at the hip, so much so that I can navigate to my Ad Settings directly from Youtube.com and Google.com and review my user profile consolidated from both platforms. You might want to put this book down so you can see your own Google profile of observed user preferences.

With this much valuable information about users, online platforms such as Google, YouTube, Amazon, Soundify, LinkedIn, Facebook, and even smaller platforms can use data to accurately push content to users.

Here are three common recommender scenarios.

1) Internet content platforms serving online content such as text-based articles and news stories, video clips, Tweets or music to users based on a prediction of user interests. Predictions might be based on common variables such as keywords, genre, artist/author/producer, and tags to content that the user has consumed in the past or what similar users like. Netflix, for example, pushes movie recommendations based on ratings provided by other users.

2) Online retail platforms showcasing purchase suggestions to users based on the user’s past actions, such as their search and purchase history, items added to their wish list, and other browsing behavior. For instance, Amazon shows product recommendations to returning users based on their previous behavior and the purchasing activity of similar users.

3) Social media platforms populating contact recommendations for individual connections, social groups, and brands/organizations based on common variables listed in your social profile and the connections and affiliations of other users in your network. LinkedIn suggests friends and industry groups based on mutual connections, keywords retrieved from your social profile, and the interests of other users on the site.

Taking into account the promising developments of augmented reality, facial recognition, 5G networks, and dedicated deep learning processing chips, there are also endless possibilities for recommender systems to follow. And whether recommender systems frighten or excite you, the best way to manage their influence and impact is to start today by understanding the architecture and algorithms that play on your personal data.

[1] Seth Godin, “Purple Cow,” Penguin House, 2005.

[2] Seth Stephens-Davidowitz, “Everybody Lies: Big Data, New Data, and What the Internet Can Tell Us About Who We Really Are,” Dey Street Boys, 2007.

Leave a Reply

Appearance
Layout
Element Style
Accent Color