VTT helps newspapers grab and hold attention of readers

The headline-spitting tool is being trialled by Kaleva Media, the publisher of over a dozen newspapers in Northern Finland.

The Finnish media company put it to the test by asking it and five experienced journalists to divide the headlines of 80 previously published articles into four categories based on their popularity among readers. The journalists were able to correctly categorise 20 but the tool as many as 50 of the articles.

“I find this to be an encouraging sign that there is definitely more scope for research,” saidHeidi Kananen, the head of data business at Kaleva Media.

“Our editors at different newspapers have already begun to use the machine on a trial basis. The Headline Machine is an exciting first step towards journalists being able to analyse the effectiveness of their headline even before the article is published.”

Although web analytics offers a convenient way to identify the most popular articles on websites, the reasons for their popularity and engagement levels remain, in most cases, unclear – particularly when it comes to the exact impact of the headline, pictures, topic, quality and positioning.

The Headline Machine was designed based on the findings of international studies and interviews with editorial staff at Kaleva Media. It uses the number of clicks and average reading time as the criteria for assigning headlines into four impact categories: ineffective, intriguing, interesting and powerful.

An intriguing headline attracts plenty of clicks but fails to hold the attention of readers. An interesting one, in turn, attracts fewer clicks but succeeds in keeping readers on the page. A powerful headline succeeds on both fronts, an ineffective one on neither.

It also takes into account factors such as the number of words and proper names, the ratio between different parts of speech, and the time and platform of publication in analysing headline options.

An effective model

A total of 7 000 previously published headlines that had been measured by means of web analytics were used to teach the tool to make its predictions, according to Sari Järvinen, a senior scientist at VTT.

The Headline Machine is based on a neural network model called BERT (Bidirectional Encoder Representations from Transformers). The model has been taught Finnish by researchers at the University of Turku.

“Thanks to BERT, the Headline Machine has learnt, for example, to recognise the coronavirus as an interesting topic this spring, despite the fact that the machine was designed on the basis of last spring’s articles. This is because the BERT model is capable of understanding the rudimentary meanings of words,” said Järvinen.

Asta Bäck, principal scientist at VTT, reminded that the popularity of headlines depends also on their topic and positioning on the page.

“Our model predicts how many clicks an article is likely to get positioned differently on a page,” she told. “We were able to see very quickly that the popularity of stories is not just about their positioning but also the topic. In our test, readers who were interested in politics were able to find the relevant articles regardless of where they were on the page, while entertainment news only got clicks in prominent places.”