Automated recommendation systems can help consumers to manage the enormous supply of online news. Most are far from transparent, however, about how and why they arrive at their recommendations or about whose interests they serve. A survey of Press Group readers found that news consumers are worried that recommendation systems may create filter bubbles and thus distort their view of the world around them. "People need to have more say in what's recommended to them," concludes KU Leuven researcher Jaron Harambam. "Greater control strengthens the individual's position in the information society and promotes healthy public debate."
According to Harambam, the provisions that online platforms make for users to influence their algorithms are generally poor. "Either you have no control, or exercising control involves a complicated set of checkboxes and sliders, as with cookie settings on websites." Against that backdrop, Harambam came up with the idea of using recommendation avatars. "When you visit a news site, you get a number of avatars to choose from, each representing a particular configuration. The avatars give the site's news algorithm a face. It's like having a friend or acquaintance to guide you around, someone you know is going to show you a certain type of content."
Workshop for news consumers
In order to further investigate the idea of recommendation avatars, Harambam and his colleagues Lawrence Van den Bogaert and David Geerts organised a workshop for fifteen Dutch and Belgian news consumers aged between twenty and fifty-six. Every day for a week leading up to the workshop, each participant was asked to complete a questionnaire after reading a news article on line. Questions included "Why did you read this article?", "What was your mood before and after reading it?" and "Was the article what you expected?"
Participants' reasons for reading news articles were then explored further in the online workshop. "A lot of the reasons were predictable, such as keeping up to date and getting background on the news," says Harambam. "Other reasons were more surprising. Wanting to see challenging or even annoying news content, for example. That led to a Devil's Advocate avatar – something we didn't envisage at the outset."
The workshop resulted in eight recommendation avatars:
Paper Boy: for topical news updates
Expert: for detailed background articles
Explorer: for articles outside the reader's normal fields of interest
Devil's Advocate: for opinion articles that provoke the reader and oppose their own world view
Diplomat: for articles with alternative ideological perspectives that expand the reader's world view
Tipster: for articles with useful information and advice
Idler: for entertaining content
Uplifter: for positive news
The Expert, the Challenger and the Relaxer
In the next phase, the eight recommendation avatars were presented to another twelve news consumers in online interviews. "The general response to the idea of recommendation avatars was very positive," says Harambam. "Interviewees envisaged recommendation avatars helping them control their news flows and improving diversity."
It was nevertheless felt that the choice of avatars was too wide, with too much overlap in some cases. In line with the feedback, the eight avatars were reduced to three: the Expert for in-depth and background information, the Challenger for articles outside the reader's normal fields of interest and ideological comfort zone, and the Relaxer for entertaining and light-hearted news.
Inspiration for editors and content developers
Like the earlier research with Press Group readers, the study found that people had fundamental concerns about the use of news recommendation algorithms, says Harambam. However, news organisations are currently looking very hard at the potential of personalisation. Harambam sees the rise of recommendation algorithms as an inevitable consequence on ongoing digitisation and the growing power of news aggregators like Apple and Google. "That makes it all the more important to take news consumers' concerns seriously and to enable them to influence recommendation algorithms," he reasons.
Harambam's work on recommendation avatars has reinforced his belief that they can help to make algorithms more transparent, more diverse in their recommendations, and more controllable. The next step is practical rollout of the avatar concept. "We want to help news platforms and editorial teams implement the idea themselves," says Harambam. He also hopes to inspire algorithm developers to give end users more control. With that aim in mind, he has submitted a paper describing his research to the Human-Computer Interaction congress CHI 21, which will be attended (virtually) by hundreds of algorithm developers.
