TTC Labs - Giving readers control over how past behaviour shapes future experience

Giving readers control over how past behaviour shapes future experience

22nd May 2019

AI is not always about the new and unusual — we can make AI trustworthy and visible by using it to reflect people’s day-to-day choices and habits back to them in an understandable way.

Product Context

Newswav participated in the TTC Labs Data Innovation Program, which was part of the first season of Startup Station Singapore in partnership with IMDA.

Newswav is a multilingual news aggregator for Malaysian readers who may want to read news from multiple sources, in multiple languages. In order to provide an accurately tailored news feed to a reader, Newswav locally and anonymously keeps track of the articles they’ve read in the app, and their stated preferences for content types and languages.

The app’s “For You” tab presents readers with an algorithmically-driven feed of articles, based on their stated preferences for news categories and languages.


Problem & Opportunity

Without affordances like an on-demand “Why am I seeing this?” feature, the logic behind algorithmically-driven feeds can be fairly opaque.

Newswav offers multiple ways of interacting with the content they present, beyond simply reading articles. Over time, Newswav could feed these interactions — reactions, comments, saves, and more specific per-article feedback — into a machine-learning model and use them to further tailor their “For You” feed to each reader.

How might we...

... clearly present the logic driving an AI-organised newsfeed to people in a way they can easily understand and update?


Team Newswav review their prototype screens

Design Features
Funneling down explanations

Tying the presence of an article in a feed directly to a reader’s recent interaction with similar articles helps explain the presence of that article without the need to fully explain the mechanics of the feed.

This is also an opportunity for interested readers to dig into the logic behind this in more detail, or, if they feel the article has been presented in error, amend the data Newswav uses to compose the “For You” feed.

Design Features
Reviewing publications and topics

If a reader’s past activity in Newswav influences the future content of their article feed, they should be able to have a high-level view of what this activity is.

Presenting a reader with a complete log of their activity would be overwhelming. Sectioning activity by publication, topic, and timeframe allows a reader to get a concise overview of what their recent activity looks like, and how it drives the content of their feed.

Next steps

As more news and social platforms move away from a “newest first” chronological presentation of content to a model where a person’s engagement, past behaviour, and larger whole-of-platform trends are used as inputs to order a content feed, the need for people to have some control and oversight of how information is presented to them is only going to grow.

Giving people a clear view of how their actions influence what they are shown feels like a useful first step for any app with an algorithmic feed. Could we also add a control layer to these interfaces, allowing people to determine which of their previous actions are used as inputs?

How might we build on Newswav’s ideas to create flexible tools and design patterns which help people understand and tailor the content of algorithmically-driven feeds, regardless of their content?

Newswav 2 next-steps-sketch