We’ll start by saying that the algorithms used to deliver content on online platforms like Facebook Trending, Google Search and Twitter are biased by definition. They prioritise and sort content based on conditions – biases – defined by their creators and operators, which would typically include topicality and personal relevance to the user.
Whilst these biases are generally designed in good faith and with a view to improving the end user’s experience, there have been a few high profile instances in which algorithms appear to be designed also to cater for the operator’s own commercial or political aims – a contentious issue which we will discuss later in the article.
How do algorithms deliver content to the user?
The finer weightings and mechanics involved in any major online platform’s content algorithm are about as closely guarded as the recipes for Willy Wonka’s fabulous confections – but there is nevertheless a great deal to be learnt from studying a platform and the information its operators make public.
Let’s take Facebook for example. Zuckerberg & Co. launched the Facebook Trending product in 2014 – if the name doesn’t ring a bell you’ll probably know it as the list of trending news topics displayed to the right-hand side of your Timeline.
When you hear the word ‘algorithm’, it’s natural to imagine a holistic, fully automated digital process. Such, however, are the nuances of content, that many content algorithms are supplemented with manual processes. Their products are editorialised.
That’s the case with Facebook Trending. An algorithm crawls both on-site data and external content from trusted sources (listed here) to produce a long list of trending topics, which are then approved and sorted manually according to Facebook Trending’s editorial guidelines. Next, a personalisation algorithm sorts the topics according to their suitability for each user – according to Facebook staff the factors involved include “the importance of the topic, Pages a person has liked, location, […]feedback provided by the user about previous Trending Topics and what’s trending across Facebook overall”.
By studying the Facebook Trending algorithm, we can gain insights on the biases or conditions that influence content delivery:
1. Buzz – to make it past Facebook Trending’s first algorithm and onto the longlist, a prospective topic will first need to generate a significant amount of engagement in bread-and-butter Facebook interactions – plus coverage on trusted external news sites. From the marketer’s perspective, this means the topic needs to be highly newsworthy, and well promoted through a combination of classic PR and Facebook marketing.
2. Relevance – for centuries we’ve been designing our own personalised content parameters by choosing which newspapers to read, which shows to attend, and more recently which TV channels to tune into. Now we have the finest minds in Silicon Valley to do the job for us. Whether that’s a good thing or not is up for debate, but the fact that Facebook Trending employs an algorithm expressly designed to personalise topic delivery for each user is a clear indication of Facebook’s faith in this novel approach. The role of relevancy in Facebook Trending allows local or special interest issues a place on relevant users’ timelines, which offers a glimmer of hope for SME marketers vying for a position in the sidebar.
3. Compliance with editorial policy – here’s where it gets interesting; where Facebook’s editorial biases (and we are not using the word pejoratively here) are applied to the results of their Trending algorithms. We would urge you to spend a while perusing Facebook’s internal review guidelines, which prescribe the diligence and editorial processes demanded of Facebook Trending staffers. We’ve picked out a few interesting points:
- Topics must represent a “real-world event” – if a trending topic cannot be proven to correlate with a single newsworthy event, it gets blacklisted. Whilst this strategy is likely effective in weeding out junk content trends, we suspect it must cause some problems in covering genuinely newsworthy trends that are not closely linked to individual newsworthy events. By focusing on topical flashpoints, we expect Facebook Trending may struggle at times to convey the bigger picture.
- Editorial staff are instructed to write for a General, PG-13 audience. The instruction refers to editorial style, but we would venture that the same philosophy would likely be applied to topic selection policy too – though that’s just a theory. It would be fair to surmise that the raw topic data gathered by the Trending algorithm has more of an ‘adult’ element to it than the final product as seen by the user.
- “Avoid defamatory allegations” – this maxim is mantra for all news publishers, and it certainly seems to have been drilled into Facebook Trending’s editorial team. Speaking anecdotally, tentative qualifiers like ‘report says’ and ‘allegedly’ appear to have been used frequently in results, though this particular trend does seem to be cooling at the time of writing. Precluding libellous statements demonstrates a bias in itself, albeit a compulsory one.
The dark side of algorithmic bias?
Constructive bias is the lifeblood of an algorithm – but how can users be certain there isn’t a commercially or politically motivated bias tucked away in its machinery that shapes the content they see?
In some cases, there might be.
In 2015 Google found themselves embroiled in a row with the EU Commission over the former’s use of search results to prominently advertise their own products and services, with listings sandwiched neatly between a single third party AdWords Ads and organic listings on the first page of results for certain searches. You can read all about it at Business Insider UK.
From an international perspective, this is something of a grey area. On the one hand you might argue that the systematic inclusion of commercially driven, Google-sponsored listings amongst third party paid and organic listings gives Google an unacceptably significant advantage over the competition. Those on the other side of the argument would say that Google have earned their prime web real estate and can do pretty well as they please.
The more you research and investigate any content delivery algorithm, the better you’ll understand the factors that make it tick and their relative importance. Whether the biases used are benign or morally dubious, they are the specification for which your content can be optimised.