WWCode Career Nav #16: Social Media Feeds and Bias
Written by Aishwarya Nair
Aishwarya Nair, Data Scientist at Trivago, shares her talk, “Social Media Feeds and Bias.” She shares her ideas about what constitutes social media and discusses differences in the motivations behind online ads. She also explains how following the crowd can change the type of ads you will see.
I work as a data scientist in Trivago in the search and ranking team. We try to create an optimal hotel ranking list for travelers worldwide. I'm very passionate about how recommendations work. We will talk about how social media feeds work, what algorithms are used behind them to show recommendations to you, and the possibility that they could also introduce some kind of societal bias for the recommendations they provide. What are social media feeds? They are in various types and forms. I do not think social media is solely photos and posts on Facebook and Instagram. It can be video recommendations on YouTube. It can also be Spotify recommendations of the next songs, what to buy next on Amazon, or even search recommendations on Google. That's feeds or recommendations you get on different social media platforms. In my investigation of how these recommendations are built, I was curious about what motivates companies to have such feeds.
Companies want to keep and retain users and have them use their products for longer. What is the intrinsic motivation for a company to have such recommendations or feeds? I looked at different social media websites or platforms with recommendations to see what kind of monetization they look at. I looked into shopping websites. Amazon or any other shopping website where you directly purchase something, the revenue model is very clear. It's a direct revenue model. When you try to buy something, their entire purpose is to recommend you something so that you purchase even more. For example, if you go to Amazon to buy a printer cartridge and you add this cartridge in your cart, you see a new recommendation to buy paper. Their intrinsic motivation is to show you recommendations to make you purchase more than required.
In the same way, there are also content streaming platforms where the revenue model is quite clear. Like with Netflix, when you buy a subscription, the recommendation of what to watch next is free of cost. There is no freemium model. There are also web content streaming platforms like Spotify or YouTube Premium, where there are parts you can see for free, certain songs that you can see for free, and then there are ads that keep interrupting you and keep coming on in between. These companies then monetize on the subscription model. They ask you to buy their subscription to stop ads.
On the contrary, with social media websites, it's not very evident how they make money. It's not very visible where the revenue comes from. Over the years, we have known that it plays like a two-sided marketplace. It's a marketing platform for many products and brands with marketing spins to advertise their products on social media. They have a targeted market where they advertise. This is why social media websites also monetize from the revenue of the ads that they display. They need to think about how much organic content versus ads to show. There is a popular saying that if you are not paying for the product, you are the product. I wanted to see what plays into this algorithm and how they decide the balance of having organic versus ad content. I found out that all these recommendation algorithms rely on user behavior. How a user behaves on these platforms, what content do they engage in, and what things do they click on? All these go into a content score and an engagement score. It then helps this recommendation system rank the content which gets maximum engagement as the top, and then ranks according to this engagement score, what you see on your feed. There are times you might have seen on your social media feeds things that are not very relevant to you but just popular.
People who spend five to 10 hours online might see a lot more ads than those who just spend half an hour. This increases the possibility of making you click on something. Social media generates revenue based on the time you spend on their website. It's a way of monetizing with you. This was a very interesting aspect to know. When I looked into the social-cultural aspect, I discovered that there is some kind of bandwagon effect that these recommendations create. A bandwagon effect is when you do something just because everybody else is doing it. This is dangerous because there is an inevitable social consequence of a majority rule. The majority decides what you see. There is also popularity bias, which reduces the quality of the content you see, showing you things that are more popular than good content. Many influencers would agree that making popular content is easier than making quality content.