2 Comments

Summary:

Facebook has reportedly done away with its once-important EdgeRank system in lieu of a system that considers about 100,000 factors in determing what content to show on users’ feeds.

news feed

The three factors that used to influence Facebook’s EdgeRank system — affinity, weight and decay — now comprise about 0.00003 percent of the total factors considered, according to a story published Friday on MarketingLand. EdgeRank was the system used to determine what shows up on users’ news feeds, but the company has replaced it with a machine-learning-based approach that takes into account about 100,000 factors.

The post goes on to list numerous new ways in which Facebook determines what to surface, including things like the type of post that it is and what type of device someone is using. Affinity, weight and decay are still important, it quotes Facebook News Feed Engineering Manager Lars Backstrom as saying, but there just a whole lot more facets and even a lot of subcategories for the original big three factors.

The MarketingLand report follows on the heels of a Facebook News Feed event earlier this month in which the company highlighted some the changes it has made to the feature.

As with the closely watched updates to Google’s PageRank system by SEO experts, the EdgeRank update is really more of a business story than it is a story about algorithms. Average users don’t particularly care where or whether stuff shows up, unless the experience starts getting annoying or starts losing value. In fact, in a world where user experience is king, it’s no surprise that companies like Google and Facebook use the latest data-analysis techniques and their ever-growing stockpiles of behavior data to predict what people will want to see.

It’s marketing types tasked with ensuring that consumers see their companies’ web pages and status updates that really care about this type of update. And, although they might openly discuss some of the changes, the companies probably aren’t too concerned about whether marketers can keep up. After all, if you want to get noticed, they have advertising products to sell.

To hear more about how Facebook uses data to improve every aspect of the site experience (and what tools it uses), check out our recent Structure Show podcast (it’s available on ITunes, too) where Barb Darrow and I interview Facebook Head of Analytics Ken Rudin.

  1. “Average users don’t particularly care where or whether stuff shows up”: Says who? And how do they know? Anyhow, I care. Besides the clumsy attempts to show me ads that match my interests, what I like least about Facebook is its opaque, capricious system for deciding which posts to show me. It isn’t even consistent across platforms. When I view my stream (or news feed or whatever they’re calling it these days) in a desktop browser and in the Android app simultaneously, I see different posts. I don’t mean the same posts in different orders; I mean some posts present on one platform are simply missing from the other. Moreover, posts come and go, seemingly at random. A post that’s visible right now may have vanished by eight hours from now (and not because the poster deleted it). Posts are often displayed wildly out of chronological order, particularly in the Android app, even though the “Sort” setting is “Most Recent.” And even when I’ve asked to see “All Updates” from certain friends, I’ve repeatedly learned from those friends that Facebook didn’t show me certain posts, and they’ve had the reciprocal experience with my posts. I would vastly prefer it if Facebook gave me the option of simply turning off all its bumbling efforts to decide what will or won’t interest me.

    Five years ago, Peter Ha opined that one of Twitter’s advantages over Facebook was as follows: “Facebook’s implementation of the News Feed doesn’t capture the full power of designing for Audience: While Twitter distributes every message consistently, Facebook decides algorithmically which update is shown to whom. Algorithmic filtering is nice in theory, but such black-box behavior is simply unpredictable for the user.” (http://tcrn.ch/adsCx9) It’s still true, and “a machine-learning-based approach that takes into account about 100,000 factors” just makes the box blacker.

    Share
    1. Derrick Harris Saturday, August 17, 2013

      Fair point. Although, it sounds like your issue isn’t that FB is using machine learning, but that it’s not doing a good job.

      Share

Comments have been disabled for this post