[Research] "Did you see my message?": Exploring how accurately phones can predict their users’ availability

florence

Tags:

By @martinpielot, associate researcher at Telefónica I+D

2 May 2014: For younger generations, not receiving a timely response to an SMS or a message is a major source of irritation, frustration and conflict. However, people cannot or do not want to be constantly tied to their phones.

But what if your phone could detect these situations and communicate them to your friends?

Here in Telefonica’s research lab we set out to explore how well phones can predict their users’ availability to messaging communication by conducting a user study which monitored the messaging behaviour and phone usage of 24 volunteers over a two week period.

The first interesting insight to emerge was that on average people see their messages within minutes. In fact, half of the messages are viewed within 6 minutes and 9 seconds.

We then went on to study how accurately the other data we logged could predict whether a notification would be seen quickly (i.e. viewed within those 6 minutes and 9 seconds). Our algorithm took into account the phone’s screen status (on or off); whether or not it was on silent or not; data from a proximity sensor; the time of day and when the user last visited the notification centre.

[youtube]http://youtu.be/gbyxOHZF1E8[/youtube]

A machine-learning model created on the basis of this data can predict (with a 70.6% degree of accuracy) whether or not a message will be seen soon after its arrival. Even if just one factor is changed, the algorithm will not be able to predict as accurately whether or not the message will be seen quickly. From the user’s perspective not all misclassifications are equally bad: it is worse if a person that is predicted to have seen the message quickly actually has not. To decrease the likelihood of this happening, we increased the level of confidence required to predict that the user will see messages fast. This means that the algorithm will predict this less often, but if it does, it will be correct in 81.2% of these cases.

This research was presented at the ACM CHI Conference on Human Factors in Computing Systems, the premier, international conference on human-computer interaction, which took place in Toronto from 26 April until 2 May 2014.

To see the full publication (Didn’t You See My Message? Predicting Reactiveness in Mobile Instant Messaging) from Martin Pielot, Rodrigo de Oliveira, Haewoon Kwak and Nuria Oliver click here.