Why Collaborative Filtering Is Putting You In a Bubble

Popular websites and internet companies, like Amazon, Netflix and Facebook, use collaborative filtering to make recommendations to users about what they may be interested in. Collaborative filtering is what lets Amazon recommend other products, Netflix recommend new shows or movies, and how Facebook determines which posts you see in your News Feed.

What is Collaborative Filtering?

This image shows an example of predicting the user’s rating using collaborative filtering. At first, people rate different items (like videos, images, games). After that, the system is making predictions about user’s rating for an item, which the user hasn’t rated yet.

Collaborative filtering is using information from many users to make predictions about the behavior or preferences of a single user. The assumption that this system uses is that if one person agrees with someone else on one issue, then they are more likely to agree on other issues as well.

More generally, collaborative filtering is using collaboration among many data sources to find patterns or reach other conclusions. It can be applied to provide recommendations for other users, or your past behavior may indicate that you are more likely do something again. Each business must apply this filtering to their own business functions to find the most valuable insights and their applications.

Some applications of collaborative filtering are ineffective due to large variances in the data they’re based on. For example, using the listening habits of millions of people to determine music tastes in a single person has a high probability of delivering poor recommendations. This is because something like taste in music is so specific to each person, and while someone may like a certain genre, that doesn’t always indicate that they like every artist within that genre.

How Does Collaborative Filtering Work?

To better understand or predict a single user’s behavior, a collaborative filtering system looks for patterns in larger group’s actions. It can be applied using a user-to-item matrix, where the system looks for connections between certain users and certain items, or with item-to-item matrices. By understanding which items are closely related to each other and which users are most likely to interact with certain items, collaborative filtering lets companies predict user behavior and make recommendations based on that data.

Read More: What is Collaborative Filtering? – UpWork

Challenges of Collaborative Filtering

    • It’s Complex and Expensive. Sorting through hundreds of thousands or millions of data points is a big process. This makes it difficult to offer real-time suggestions online. For a company like Amazon with millions of users, scaling a collaborative filtering system can require large budgets and massive computing resources.
    • Data Isn’t Always Complete. Many times the data that these filtering systems use doesn’t accurately explain a users’ interests or intentions. For example, just because a user watches a video doesn’t mean that they liked it, or someone purchasing a
    • It Needs Historical Data. Another roadblock that these filtering systems must overcome is that they need data to function. This makes it impossible to recommend new content to users. It’s also difficult to make recommendations to users who have just started using a particular product or service. This leads to a bias towards older, already popular items. As a result the popular items get recommended more, even if a newer addition may provide a better match.
    • Failure to Account for Fringe Users. Certain users in these systems will have views that are contradictory to what the system has learned. For example, if a shopping site notices that most people who buy peanut butter also buy jelly, but you like eating peanut butter with onions and not jelly, the system won’t provide a valuable recommendation for you.

Collaborative Filtering and Filter Bubbles

One of the problems with filtering and content recommendation is that these lead to filter bubbles. When an algorithm is determining which content a user sees, people tend to only be presented with information similar to what they have already looked at. This reinforces peoples views and opinions rather than providing an objective look at a given topic. Filtering content makes internet tools feel user-friendly and convenient because everything seems to “fit” your interests. Ultimately though, it’s keeping you trapped in a bubble of just a portion of the entire range of information available on the internet.

Read More: What Are Filter Bubbles? How To Avoid Them?

As filtering systems continue to improve, they will become more context-aware and better at matching users with information, products and content. This makes systems more convenient for users, but isolates them further in their own filter bubbles. It’s easy to see that in the not-so-distant future, these suggestion engines and recommendations could be so ingrained in our internet use that it could be nearly impossible to find information that contradicts with our views, even if the alternatives are mainstream viewpoints.

Conclusion: Filtering Is User-Friendly, But It’s Risky

It’s hard to argue that content filtering and algorithmic content feeds are anything but user-friendly. They provide access to better content without having to sort through it all on your own. There are countless applications of these content delivery systems on the internet. Social media sites are the most notable example. It’s worrying though that as these systems improve they could lead to better information access for the masses, or they could end up trapping internet users in even more concentrated bubbles than already exist.

Read More: