Large-Scale Evaluation of Call-Availability Prediction (UbiComp ’14)

Roughly 1/3 of all phones calls are not picked up. With this work, we explored whether the called phone can know in advance, whether its user is likely to pick up a call. This would allow to, amongst other things, communicate (non)availability in advance to the call or trigger intelligent muting.

This work shows that mobile phones can predict with an accuracy of 83.2% whether its user will accept an incoming phone call or not. When personalizing those models, the accuracy can be increased to 87%.Therefore, the phone needs to keep track of 15 features, such as the time since the last call, the day of the week, or the ringer mode. The 5 strongest predictors are:

(1) time since the last ringer mode change,
(2) time since the screen was last turned on or off,
(3) screen status (on/off),
(4) time since the phone was last (un)plugged, and
(5) time since the last call.

These findings show that it is possible to create an automated availability status for phone calls. Integrated into any phone call application, it could help to manage expectations by sharing the availability prediction with potential callers, and through that greatly impact the overall user experience. Further, knowing whether is a user is likely to take a call might be useful to intelligently allocate resources in a multi-device messenger environment.

To obtain the necessary data, we instrumented a previously-developed application called Silencer with anonymous data-collection facilities. During a two-month period, the app logged how 418 users reacted to 31311 phone calls. Alongside each call, the above mentioned 15 features were collected. Using a Random Forest, we computed the accuracy of a generic model, of personalized models with different numbers of calls (in average, 50 or more calls are needed to outperform the generic model – so it should be very quick to generate accurate personalized models), and to determine the prediction strength of each feature.

This research is described in detail in the paper Large-Scale Study of Call-Availability Prediction, which will be presented at the ACM International Joint Conference on Pervasive and Ubiquitous Computing, held in September 2014 in Seattle, USA.

Share this:

The fallacy of WhatsApp’s “last seen” status

Last Seen = Fast Response?


When sending a message with WhatsApp, senders often check the receivers “last seen” status to judge whether the message will be read soon. It shows when the receiver had last openend the application.

it gives me a timeframe and allows me to estimate when my message will be read

Intuitively, if the receiver was online only recently, s/he is likely to be near the phone and see the message soon.

However, results from our recent study on predicting how fast people attend to message notifications indicates that “last seen” is almost as weak as a random guess.

How fast people view WhatsApp messages

We installed an app on the phones of 24 volunteers, which logged, amongst other things, each time that

  • WhatsApp is opened or closed,
  • a WhatsApp message is received, and
  • the user sees the WhatsApp message, either in the notification drawer or in the app

For these volunteers, the median delay between receiving and seeing a WhatsApp was 7.81 minutes, i.e. half of the messages were viewed within 7.81 minutes and the other half later.

We used this time to split the data set into two parts: fast = seen within 7.81 min, slow = seen after 7.81 min. This means, a random guess whether a users sees then message fast or slow has a chance of 50% to be correct.

Not much better than random guess

Next, we used the log data to train a state-of-the-art machine-learning model. We checked how well “last seen” allows it to predict whether the message is seen fast or slow.

It turned out that the prediction was correct in 58.8% of the cases — only 8.8% better than the random guess.

Do not overly rely on “last seen”

Of course, this study has its limitations. The 24 volunteers were in their late twenties and early thirties. Other demographics might exhibit different behavior.

However, the results indicate that we should not overly rely on “last seen” when we want to estimate the availability of our friends.

Share this:

Didn’t you see my message?! (CHI ’14)

“Didn’t you see my message?!”

For the younger generations, not receiving a timely response to a SMS or message is a major source of irritation and frustration.

However, people cannot or do not want to always attend to their phones all the time.

What if your phone would infer these situations and communicate them to your friends?

Only how would the phone know?

Our research at Telefonica Research shows that these predictions can be done by simply monitoring a phone’s screen status (on/off), ringer mode, proximity sensor, the hour of the day, and when the user last visited the notification center.

In a user study, where we tried the system with 24 participants over 2 weeks, we learned that half of the messages are viewed within 6.15 minutes, and the other half after that.

A machine-learning model created on the basis of this data can predict with an accuracy of 70.6% whether a message will be viewed within 6 minutes or later. If the prediction is that the message is going to be viewed within those 6.15 minutes, it is even more conservative: the precision of the model is 81.2% in this case.

This research will be presented at the ACM CHI Conference on Human Factors in Computing Systems, held in Toronto, Canada in May 2014.

Martin Pielot, Rodrigo de Oliveira, Haewoon Kwak, Nuria Oliver
Didn’t You See My Message? Predicting Reactiveness in Mobile Instant Messaging
Proc. CHI ’14 Conference on Human Factors in Computing Systems, ACM, 2014.

Share this:

Telefonica Research at CHI ’14

Telefonica Research will be represented with 2 full papers and 2 ToCHI articles at ACM CHI ’14, the premier international conference of Human-Computer Interaction.

Didn’t You See My Message?

Martin Pielot, Rodrigo de Oliveira, Haewoon Kwak, Nuria Oliver

We found that monitoring the phone (screen activity, notification center access, proximity sensor, ringer mode) allows to predict whether a person will attend to a received message fast or not (pdf).

A brief but more detailed description can be found in in more recent blog post.

Large-scale assessment of mobile notifications

Alireza Sahami Shirazi, Niels Henze, Martin Pielot, Dominik Weber, Albrecht Schmidt

As part of the study, we published an Android app on Google Play that forwards all phone notifications to the browser (via plugin). More than 40,000 people thought this was a brilliant idea and downloaded the app. We used the app as a vehicle to log and analyze all notifications that users receive (pdf).

A Large-scale Study of Daily Information Needs

Karen Church, Mauro Cherubini, Nuria Oliver

My colleagues have conducted one of the most comprehensive studies of information needs to date. For three months, they probed information needs via experience sampling and daily diaries, to understand “the types of needs that occur from day to day, how those needs are addressed and how contextual and demographic factors impact on those needs” (details on Karen’s website.)

Influence of Personality on Satisfaction with Mobile Phone Services

Rodrigo de Oliveira, Mauro Cherubini, Nuria Oliver

My colleagues connected the phone use habits of 603 volunteers with personality traits and customer satisfaction, and found that “(1) extroversion, conscientiousness, and intellect have a significant impact on customer satisfaction—positively for the first two traits and negatively for the latter; (2) extroversion positively influences mobile phone usage; and (3) extroversion and conscientiousness positively influence the users’ perceived usability of mobile services” (ACM Digital Library).

Share this:

Open CSV files by double click with Excel 2011 (OS X)

Excel can open and interpret CSV files on double click.

However, for some locales it may happen that all the content appears in the first row.

This is how I fixed it.

Clean up Preferred Languages

In my case, I am using Excel 2011 for OS X 10.9. The CSV was not interpreted correctly, because German appeared in the Preferred languages of the Language & Region settings.

The simple fix was to remove all preferred languages except from English (United States) so that it would become the primary language, and then re-add the other languages.

After that, Excel 2011 flawlessly opened a comma-separated CSV file on double click and distributed the values correctly over the rows.


German, and other languages, use the comma instead of the dot as decimal separator (e.g. 123,45 instead of 123.45). Thus, Excel on German machines expects semicolons to separate values in a CSV file. The fix above will help to work with CSV files in international environments.

Share this:

Ambient Timer – Unobtrusively Reminding Users of Upcoming Tasks with Ambient Light

Alice is working on a report for the head of her department. At the same time, there is a meeting scheduled in thirty minutes, which she has to attend. The Ambient Timer is already illuminating the wall behind her monitor in a low-attention state, so Alice feels confident that she will be reminded of the meeting. A few minutes before the meeting, the status of the ambient light display has changed to a more salient, intense output. While she is still working on her report, she slowly becomes aware of the nearing deadline and starts finishing the paragraph she is currently working on. One minute before the meeting the light has become so salient that it is hard to ignore. Alice stores the document on the server, puts her computer into sleep mode, and arrives at the meeting on time.


A timer that uses ambient light

The Ambient Timer is a research prototype developed in the Interactive Systems Group of the OFFIS Institute for Information Technology. Its goal is to gently remind information workers about upcoming events. such as illustrated in the scenario above. It uses LED glued to the back of the monitor to illuminate the wall in the peripheral field of vision of the worker.

User Study

With this prototype, we conducted a user study in collaboration with the HCI and Mobile Computing Group of Telefónica Research. We experimentally studied two instances of the Ambient Timer:

  • expo: a gradual change from green to red, becoming exponentially faster, and
  • sinus: a sinusoidal change between red and green which became increasingly faster.

We compared these reminders against two traditional techniques to keep track of appointments:

  • a clock, such as the one in the corner of your computer screen, and
  • a popup alarm, such as when you use Outlook, Lotus Notes, or the OS X Calendar for your appointments.

For the study, we asked participants to copy and correct texts. Meanwhile, a 10-minute timer was running in the background. The task was to finish as many texts as possible in 10 minutes, but without “overshooting”, i.e. having an unfinished text after 10 minutes. In the expo, sinus, and clock conditions, the remaining time was presented by the Ambient Timer or a clock, respectively. In the popup condition, no time was given, but a popup informed the participants 30 seconds before the end of the time limit.

The experiment used a repeated-measures design, i.e. each participant tested each of the four reminder systems in counter-balanced order.


Our results show that participants experienced significantly fewer interruptions when using Ambient Timer in the expo condition, i.e. with an exponential change from green to red, compared to all other reminder techniques in our experiment. Their average typing speed was significantly faster when in this condition, too. Participants ranked this design best, felt most confident using it and preferred it over all other techniques.


This experiment shows that using light in the periphery around the monitor is a great way to provide information workers with information in an ambient way. Used as Ambient Reminder, ambient light might help to structure typical office work, which is often a mix of concentrated desktop work and scheduled meetings and appointments. It allows office worker to avoid to constantly check the clock or be interrupted by alarming popups interrupt.


The details of the experiment have been published in the 14th IFIP TC13 Conference on Human-Computer Interaction, held in September 2013 in Cape Town, South Africa:

Heiko Müller, Anastasia Kazakova, Martin Pielot, Wilko Heuten and Susanne Boll.
Ambient Reminder: Unobtrusively Reminding Users of Upcoming Tasks with Ambient Light.
INTERACT ’13: 14th IFIP TC13 Conference on Human-Computer Interaction, 2013.

Share this:

Peripheral Vibro-Tactile Displays

If you are sitting, which parts of your body are currently touching the chair? Are you leaning on a backrest or an armrest?

Think about it! Now, you are aware! But have you been just before reading this post? Probably not!

This is the beauty of peripheral perception, i.e. perceiving sensory input in the periphery of our attention. Your brain perfectly knows how to process the touch input it gets from the different parts of your body so that you do not fall of the chair. At the same time, you can perfectly focus on reading this text.

Using our Sense of Touch for Periperphal Communication

But, could this property of our sense of touch also be used to communicate information in the periphery of attention via mobile and ubiquitous computing devices? For example, imagine a bracelet indicating the time remaining until your next appointment, or your mobile phone indicating that there are no unread emails, messages, or social network updates to attend to.

Peripheral Vibro-Tactile Displays

In our research on peripheral, vibro-tactile displays, we made first investigations to prove that such information presentation could be possible with vibration motors, or vibro-tactile displays, as they can be commonly found in our mobile phones.

Study: Exposing People to a Constant Heartbeat

15 participants wore a vibro-tactile display in their pocket for 3 days. The display was set to create a constant, soothing, heartbeat-like vibration pattern. Via mobile phone, the participants adjusted the intensity, so that the vibration was barely perceptible.

Death Events: Testing Awareness

To test whether the vibration was still perceived, it died after 15 to 60 minutes. As soon as the participants noticed, they had to acknowledge the death of the vibration by pressing a button on a mobile phone. In the study, the majority of the death events were noticed between 1 and 10 minutes after the vibration had died. This is an indicator that participants were still aware about the vibration, even though it was set to very low intensities.

Testing Ambientness

To check whether the vibration had left the participants conscious perception, i.e. the focus of attention, we opened a questionnaire on the phone once the participants had pressed the button. In 67.7% of the cases, the participants indicated that the subjectively did not think that they had noticed quickly that the vibration had died. Additionally, in 94.4% of the cases, the participants reported to not be annoyed by the vibration. These two results indicate that the heartbeat vibration was indeed not in the focus, but in the periphery of attention.


These results provide first evidence that vibration patterns can form non-annoying, lightweight information displays, which can be consumed at the periphery of a user’s attention.

However, these findings are only first steps. We need more evidence to back up the findings, and we need more insights into how to adjust the intensity of the vibration pattern to different situations, so that we always hit the sweet spot of being just barely perceptible.


The details of this study will be presented at ACM MobileHCI ’13, the 15th International Conference on Human-Computer Interaction with Mobile Devices and Services, held in August 2013 in Munich, Germany.

Martin Pielot and Rodrigo de Oliveira.
Peripheral Vibro-Tactile Displays.
MobileHCI ’13: 15th International Conference on Human-Computer Interaction with Mobile Devices and Services, 2013.

Share this:

I Never get the Good One’s! How many Paper to Review?

I never get the good papers for review.

If this thought has ever crossed your mind, you are probably not alone. Good conferences in the HCI field typically accept only 20-25% of the papers submitted for review.

So how many papers to accept for review

… to review at least one good one?

It may seem obvious: the number should depend on the acceptance rate of the conference: 4 papers for 25%, 5 papers for 20%, 6 papers for 16.6%.

So, you are doing this already, and you still seem to get only the to-be-rejected ones?

This is, because probability computation does not always follow the intuitive approach.

Compute by rejection rate

The key is to compute by rejection rates and multiply them per paper.

If you review one paper from a conference with 20% acceptance rate, it’s likelihood to be rejected is 80%.
For two papers from the same conference, the likelihood that both are rejected is 80% * 80% = 64%. (not 60%, as what our intuition might tell us)

The row continues:

  • three papers = 51.2%
  • four papers = 41.0%
  • five papers = 32.8%

So, even if you review five papers in this conference, the likelihood is 32.8% that all of them will be rejected.
See the diagram below for different acceptance rates (25%, 20%, 15%).

Likelihood of Reviewing Accepted Paper by Acceptance Rate of a Conference
Likelihood of Reviewing Accepted Paper by Acceptance Rate of a Conference


So, how many papers to accept for review?

If you want to have a 20% 80% chance of reviewing at least one accepted paper, you have to accept the following number of papers for review:

  • 6 papers for a venue with a 25% acceptance rate
  • 7-8 papers for a venue with a 20% acceptance rate
  • 10 papers for a venue with a 15% acceptance rate


Share this:

Use 18pt Font Size for Readers with Dyslexia

Dyslexia: a common reading disability

Dyslexia is a neurological reading disability, which impairs a person’s ability to read and write. In the media, we often hear about dyslexia as a gift in the context of famous people, such as Steve Jobs. However, in reality, depending on the language, a significant chunk of the people suffer from dyslexia, e.g. 10 to 17.5% in the US. For most of these, dyslexia is not a gift: the most common way of identifying dyslexia in children is bad performance in our reading-centric education system.

Can the right presentation parameters improve reading?

The good news is that reading increasingly takes place via electronic displays, where we can adapt the presentation of text to make it easier to read for people with dyslexia. Therefore, led by Luz, we (Luz Rello, Martin Pielot, Mari-Carmen Marcos, and Roberto Carlini) set out to find optimal values for the most simple parameters of presentation: font size and line spacing.

Eye-tracking study exploring font size and line spacing

The study was conducted by Luz Rello in the Universitat Pompeu Fabra (UPF) in Barcelona, Spain. 28 people (15f, 13m), aged 14-38, with a confirmed diagnosis of dyslexia took part in the study. They were asked to read Wikipedia articles that were presented with different font sizes and line spacings. The study used eye tracking and questionnaires to measure readability and comprehension.

The experiment compared:

  • Font sizes: 10, 12, 14, 18, 22, and 26 pt.
  • Line spacings: 0.8, 1.0, 1.4, and 1.8.


To make a long story short, line spacing did not have much of an impact. Only 1.8 line spacing lead to worse comprehension compared to 0.8 line spacing.

Regarding font size, however, the results were surprising. When we look for optimal font size in the web, we either find soft recommendations, such as “allow to adjust
or values around 12pt / 14pt.

However, our results provide strong evidence that for people with dyslexia, readability and comprehensibility of a text increases with font size, which an optimum around 18pt.

In particular, we found that:

  • The objective readability, which is indicated by the fixation duration recorded with the eye-tracker, steadily increased until 18pt.
  • The subjective readability was highest for 18pt and 22pt.
  • The subjective comprehensibility was highest for the three largest fonts: 18pt, 22pt, 26pt.

Conclusions: use 18pt font size for your website

Hence, when designing a website that shall be friendly to readers with dyslexia (remember, 10-17.5% of the population!), use large fonts. Since there was no improvement at larger font sizes, 18 pt font size hits the sweet spot.

Complete report

The complete scientific report can be found below.

Luz Rello, Martin Pielot, Mari-Carmen Marcos and Roberto Carlini.
Size Matters (Spacing not): 18 Points for a Dyslexic-friendly Wikipedia.
W4A ’13: 10th International Cross-Disciplinary Conference on Web Accessibility, 2013.

This work was published at the 10th International Cross-Disciplinary Conference on Web Accessibility, held 13-15th May 2013 in Rio de Janeiro, Brazil.


Share this:

How the Phone’s Vibration Alarm can help to Save Battery

Not sure how long my hero’s battery will last with GPS on and my phone vibrating every second to indicate if on right track!?!

– This and similar concerns have frequently been expressed when I presented the PocketNavigator – a navigation system guiding pedestrians by vibration patterns instead of spoken turning instructions.

To quantify how much battery power is actually lost to constantly repeated vibration pulses, I tested the battery consumption of two different patterns in comparison to a non-vibrating phone.

In brief, in my setup, the vibration cost less than 5% of the battery life. As comparison: leaving the screen on will drain the phone’s battery in 2-3 hours. In consequence, instead of draining the battery fast, vibration can even help to save battery if it allows users to leave the screen turned off.

Test Configuration

The apparatus created heartbeat-like vibration patterns, i.e. patters consisting of two pulses followed by a long pause. The apparatus was run three times. Each run used a different pulse length, i.e. 30 ms, 60 ms, and 0 ms (no vibration as baseline).


The following diagrams show the remaining battery as it changed while the app was running.

The battery lasted

  • 24.71 hours for 0 ms pulse length (baseline)
  • 23.48 hours for 30 ms pulse lengths = 95.0 % of the baseline, and
  • 23.48 hours for 60 ms pulse lengths = 95.0 % of the baseline.

Using linear approximation to account for the fact that the battery was never 100% charged when the trials commenced, we also calculated the trend lines (see Diagrams, used Excel’s linear approximation), which changes the prediction to

  • 24.18 hours for 0 ms pulse length (baseline)
  • 23.28 hours for 30 ms pulse lengths = 96.3 % of the baseline, and
  • 23.60 hours for 60 ms pulse lengths = 97.6 % of the baseline.


Battery life in all cases was around 24 hours, sufficient for normal use. Constant vibration reduced battery life by 2.4 – 5.0 % minutes. Increasing the vibration length from 30 to 60ms per vibration pulse had no effect on battery life. As comparison, when the screen is constantly kept on, the battery drains within about 2-3 hours.

Hence, the additional battery loss is justifiable when considering that at the same time we gain the ability to continuously communicate information to the user. When using short vibration pulses, desigers do not even have to consider the effect of the pulses’ lenghts on battery life.

Take Away

This data shows that the impact of having the phone emitting vibration pulses constantly is not very high.

This means that as means to constantly convey information, e.g. as navigation system that is supposed to convey information all the time, vibration has a much lower impact on battery life compared to the screen, which empties the battery in a few hours. On a Nexus One, vibration can allow to constantly convey information for almost 24 hours, enough for the typical smartphone user who has gotten used to charge the phone every night.

Share this: