Andreas Weigend | Social Data Revolution | Fall 2014
School of Information | University of California at Berkeley | INFO 290A-03
Audio: weigend_ischool2014_7a.mp3 and weigend_ischool2014_7b.mp3
Transcripts: weigend_ischool2014_7a.docx and weigend_ischool2014_7b.docx
(not done yet)


Fitness can be looked at from two different perspectives. The first is that of health, well-being, and fitness. The second perspective is that of the business equation of fitness, which is still under development at the moment. Today, we will discuss both.

Special Opportunity Guest - Firechat

Before we get into the fitness aspect of the class we first had a special opportunity guest - the founder of Firechat - Micha Benoliel.

What was the philosophy behind Firechat?

The initial purpose behind Firechat was to help people connect during events such as Burning Man or music festivals where your network might be congested or, in the case of Burning Man, where there isn't a lot of signal. Firechat is an off-the-grid instant messaging application that makes use of peer to peer mesh networking. It leverages the radios of the phone (without necessarily connecting to a cell tower or a wifi hotspot) to other phones nearby. Communities can build their own networks - which is what happened in Hong Kong when more than .5 million people installed Firechat in a week. Joshua Wong, afraid that the government would shut down access to the internet during protests, told all of his friends to download Firechat. Firechat still works even if mobile networks are congested by relaying to people around based on smartphones that become internet nodes. There is now more and more interest for the potential application of the technology to disaster recovery situations.

The Social Data Revolution of Firechat

One key application of social data and the revolution right now is happening in Hong Kong. On the one hand, we can say that the mainland Chinese government cannot see the messages in their central server and thus might not be able to triangulate where individual comments originate from. This is a big illusion in the side of the user - this idea that there is anonymity in Firechat. The way that Firechat works (namely via proximity signal location) the government can send an agent within seconds to someone - it is much more efficient because there is this location built in on the signal strength wave. In theory with working technology Firechat would be great for big events like Gay Pride but it is limited in terms of protecting the user.


Sensors have become very cheap in recent years. An accelerometer, for example, can cost as little as $0.72. Now there exist people based sensors, which collect information about ourselves. This is called the ‘quantified self’. These manifest sometimes as what we referred to in class as "wearables"- a device that one supposedly wears for extended periods of time that will collect data. This could be anything from bracelets that act as a pedometer to an anklet your baby wears to monitor it's sleeping pattern. The example we discussed in class is Pebble (, the developer of the Pebble smartwatch, which creates a customizable experience for each user and offers the sensors with which to create apps to collect information on the quantified self. Today Susan will join us and will discuss the challenges Pebble faces with extracting information from the rich data pool they have collected.

What is Pebble?

  • The pebble is a smart wearable or smartwatch.
  • One of the most successful Kickstarter campaign of all times (10 million dollars from over 60,000 people).

The special thing about the Pebble smartwatch is that it is open to third party developers to build applications for the Pebble. Examples of applications of the Pebble are: fitness applications using the motion tracking functionality of the Pebble (step tracking, running and swimming motions), applications which update users on sport’s scores, and many more. It makes a really customizable experience for consumers which means that the data that Pebble gets is very rich too.

Pebble Misfit app

Developing new business cases

How to understand people’s usage and engagement over time?

Social gaming industry

Analogy with the social gaming industry:
  • Highly developed data analysis vocabulary to understand user engagement.
  • Tracking daily active user accounts and user retention, which are measured in a standardized way

These kind of measures are usable for three key reasons:
  • Minimal variation of engagement in video games.
  • The product life cycle of social games is short and stops abruptly.
  • Users react instantaneously to changes that the software developers implement.

It makes sense to track these changes, because it is actionable for executives in the social gaming industry. Unfortunately, this is not possible with wearables.

The measures are not usable in the wearable industry, because:
  • The product life cycle for wearables is longer.
  • The Pebble provides more value which makes users become more invested in it.
  • There is no abrupt stop in usage.
  • There is a high level of variety in terms of engagement styles (on an individual level and an aggregate level).

Evaluating the user experience

How to interpret a good user experience, and how can Pebble as a company find out what they are successful at and what they need to do better?
Initial equation: Pebble happiness (Φ) = ∑ (user * hours of engagement) / total number of users in pool). What is the problem with this equation?
  • The outcome is very hard to match with reality, it has no meaning, so you cannot draw actionable conclusions from it.
  • It is difficult to communicate to executives.

What kind usage patterns are there and how do they look like? How can Pebble start to get a sense for when people are actually engaging with the watch or are just passively using it?
  • Using the accelerometer which is incorporated in the device.
  • It is highly sensitive and tracks in 3 dimensions.
  • Developers can program different gestures in the applications they develop.
  • The downside of the accelerometer is that it also tracks a lot of noise data.

So how can Pebble separate the noise from the actual engagement by the user?
  • A phase transition (PT) was taking place. This means it is going through some qualitative change (state 1 à state 2).
  • For example, it could a noise cut off (above cutoff is meaningful human motion).

They discovered a pattern: some people wore the watch during their sleep. This particular discovery opened up a new business case for Pebble. They could now recommend users who were wearing the watch 24 hours a day, but did not use a sleep application, to start using a sleep application. These kinds of patterns are interesting, but do not provide the full story. If Pebble would be trying to optimize the hours of use of the watch, they would not be capturing the quality of the user experience.

Future of Pebble

Where stands Pebble now? They recognize that they still need to fill in missing parts of the puzzle.
  • For example questions like, if you bought the watch 6 months ago, how likely is it that you are using it today?
  • Are you still using it regularly or have your engagement patterns changed over time?

Pebble is thinking more intuitively and more deeply about what the user experience is and what they can measure to acquire information about what is good for the users and what not.


The use of data science can also be applied to areas where the consequences are more severe.

Swallowing disorders

Swallowing disorders (also known as dysphagia) is a serious and often overlooked problem
  • Affects 30-40% of American population over 65
  • Direct of indirect cause of hundreds of thousands of deaths each year

Eating strategies

To minimize the risk, people get assigned a set of strategies to follow
  • Limited number – around 15
  • Example: tucking in your chin when swallowing food (see picture)



People often forget to comply to these strategies
Wanted: A friendly, but vigilant caregiver
  • Not an easy task for a human
  • Goes against social conventions to constantly correct people’s behavior

Easy task for machine
  • If the rules are well-defined, they will always be followed
  • BUT: Needs some form of input

Computer vision

In recent years computer vision has come a very long way, see for instance Kinect. The Kinect uses special camera technology for depth detection, but it is also possible to understand a great deal about what a regular camera sees through image recognition.
  • Look for defined visual landmarks
  • Compare to 3D face model to detect orientation of face
  • Train system to recognize compliance with strategies

With this technology, the patient can simply have an iPad in front of her that detects and issues a warning every time she forgets to tuck in her chin.

Data collection

The system would help doctors keep track of how well their patient is doing
  • Does she comply with the assigned strategies?
  • Does it seem to help?
  • Is there a correlation between non-compliance and issues?

This information has so far been unavailable to the medical staff.
Furthermore they would get very useful information in a general sense
  • How people many are complying?
  • How does it relate to their treatment?
  • Are they taking their medicine?
  • Does the medicine seem to work?

The point about medicine are especially interesting to the medical world, since traditional medical studies are very expensive and doesn’t always give a very precise depiction of reality. If they instead could get contiguous and real data through modern technology, it would be very valuable to them.

Question for thought: How else could data science help impact the fields of health and medicine? This example from class is quite limited in scope (and probably one of the more unknown applications of data science for health.

This page initially created by (3 students total):
  1. Yujin Lee (
  2. Tom van Norden (
  3. Jeppe Stougaard (