Dwell time a proxy for engagement

9 October 2014

SUNNYVALE, CA: The idea of measuring how long users spend with content as a more reliable indicator of engagement than click-based signals is gaining momentum.

Scientists at Yahoo! Labs analysed user interaction data from traffic on the stream on the Yahoo home page and concluded that what they refer to as 'dwell time' – other advocates of this metric call it 'engaged time' or 'attention minutes' – could be a useful proxy to user satisfaction for recommended content.

The Yahoo researchers reported that their findings matched what one might intuitively suppose, namely that users have less dwell time per article on mobile or tablet devices than on desktops and that users dwell more on longer articles, up to 1,000 words.

Users were also likely to spend the longest time on articles about politics or science and the least on articles about food or entertainment.

But, noting that different users exhibited different content consumption behaviours even for the same piece of content on the same device, Yahoo devised the idea of 'normalised user dwell time' in order to remove variances due to the differences in the context.

The scientists reported that when they used the normalised dwell time instead of clicks as the optimisation target in recommendation algorithms, performance improved.

"Content recommendation models perform better when using dwell based metrics," they concluded, although they could not say if these also drove long-term user engagement.

The validity of such metrics was given an extra boost recently when the Media Ratings Council accredited the 'attention minutes' measurement developed by Chartbeat, a real-time analytics business.

CEO Tony Haile observed to GigaOm that people had long liked the idea "but now it's official … There can be an attention economy, in which both publishers and advertisers buy and sell attention minutes or metrics as a measurement".

He noted that time was the only unit of scarcity on the web. "Just like any economy of scarcity, anyone who captures most of it can charge more," he told Advertising Age.

Earlier this year the Financial Times ran a trial around this idea as it explored how to sell display ads based on how long the audience spent with its content.

Data sourced from Marketing Land, GigaOm, Advertising Age; additional content by Warc staff