For Your Eyes Only: Consuming vs. Sharing Content

@SNOW/WWW, 2016, by Roy Sasson and Ram Meshulam

fyeo

Do you share on Facebook every page that you visit?

Assuming that the answer is “no”, how do you determine what to share? The answer to this question is meaningful for publishers, content marketers and researchers alike. Many of them try to infer user engagement from the sharing activity of users, among other signals. The underlying assumption is that highly shared articles are highly interesting/engaging for the users.

Based on more than billion data points from hundreds of publishers which use Outbrain’s Engage platform worldwide, we show that the above assumption is not necessarily true. There is a dissonance between what users choose to read in private vs. what they choose to share on Facebook. We denote (log of) the ratio between private engagement (measured by click-through rate) and social engagement (measured by share-rate) as the private-social dissonance.

The private-social dissonance consistently varies across content categories. Content categories such as Sex, Crime and Celebrities are characterized by a high positive dissonance. Articles under these categories tend to be visited relatively more than being shared. On the other hand, content categories such as Books, Wine and Careers are characterized by a negative dissonance. Articles under these categories tend to be shared relatively more than their popularity.

dissonance

This figure shows content categories, sorted in a descending order by their private-social dissonance. To the human eye, inspecting the categories from top-left to bottom-right the picture is clear. Users tend to read without sharing articles from categories that could harm (or not increase) their social appeal. On the other hand, users tend to share categories that are not relatively popular, yet they reflect a positive and socially desirable identity of the sharing user. Our results are time consistent and did not vary substantially during a period of one year.

To further test the value of social signals in terms of engagement, a model which utilizes different signals and produces click-prediction was trained and deployed on a live recommendation system. The resulting weights ranked the social signal lower than other signals, such as click-through rate.

What next?

It would be interesting to investigate the relation between private-engagement and Facebook’s recently announced Reaction buttons. Will some of the new buttons have a close-to-zero dissonance, and thus can be used as an accurate metric for engagement? Twitter is also a good candidate for investigation. Another direction is to use refined private-engagement signals instead of CTR, such as time-on-page or scrolling behavior. An interesting question can be – ‘do users actually read what they share?’.

Another interesting direction is to utilize the private-social dissonance in a classifier for inappropriate content. Articles with high positive dissonance are many times inappropriate to some extent. Such a classifier is based on users’ behavior and does not rely on natural language processing or image processing.

In conclusion, publishers, marketers, architects of recommendation-systems and anyone who uses social signals as an engagement metric should be aware of the private-social dissonance.

For more details, please refer to our paper entitled: “For Your Eyes Only: Consuming vs. Sharing Content” by Roy Sasson and Ram Meshulam, Third Workshop on Social News on the Web @ WWW ’16 (SNOW 2016), Montreal, Canada, April 2016.