The AppSensor that we provide in this project is a virtual sensor for measuring mobile application usage on the Android plattform.

The AppSensor is collecting the following four types of events on mobile application usage:

• App installation
• App updates
• App removals
• App being used (determined by events open and close)

The user’s application usage is sampled as a sequence of events, i.e. every usage of an app results in one event; a user’s swapping between two or more apps also results in a list of events. In the database of a central server an event has the following attributes:

• Type of event
• Package name of the application
• Start time of event (UTC timestamp)
• Offset from local time zone to UTC (in hours)
• Usage time (milliseconds)
• Information of location API ** Longitude ** Latitude ** Accuracy ** Altitude ** Speed
• Powerstate (connected / unconnected to charger)
• Powerlevel (battery load in percentage)
• Time of last screen on for sessions (UTC timestamp)
• State of headphones (plugged in: yes / no)
• Orientation of device (portrait / landscape)
• WiFi state (turned off / turned on / connected)
• Bluetooth state (turned off / turned on / connected)
• GPS state [implementation planned]
• Identification of device ** Hash of IMEI device id (deprecated, unique per device) ** Installation id (unique per installation) ** Information about device
• Model name
• Screen resolution o Android API Level ** Version of the AppSensor library being used (form AndroidManifest)
• Client IP4 address (only through server script)

The AppSensor is currently being sampled with a frequency of 2Hz, i.e. every 500ms we capture which application is currently running. Whenever we observe a different application, we create a new event of type app usage for the current app. The AppSensor starts observing app usage when device is unlocked / waked up from stand-by / screen is turned on.

To save battery and for better performance, this real-time data is saved in memory. Most of the attributes of an event are determined when an application is started (e.g. package name, type, location). Some pieces of information are constantly being used to update the event, e.g. usage time and device orientation.

Every time the device goes into stand-by mode, i.e. when the screen turns off, we write all data from memory into a local MySQLite database on the Android device. Further, the AppSensor stops sampling app usage since the user cannot use any app when the device is locked.

Synchronization with the server is done via the Android sync manager. The data is send as a CSV file, where all redundant fields are blanked to save bandwidth. On the server side, a PHP script receives the CSV file as HTTP POST. It reads the file and writes all data fields into a dedicated table of a MySQL database. Before uploading, such values that are constant but device-specific are added to the event, e.g. screen resolution or device model.

If you are using the AppSensor for running research studies and think about publishing your results, you should cite the following paper as a reference for AppSensor:

• Matthias Böhmer, Brent Hecht, Johannes Schöning, Antonio Krüger, Gernot Bauer: Falling Asleep with Angry Birds, Facebook and Kindle – A Large Scale Study on Mobile Application Usage. In: Proceedings of Mobile HCI 2011. Stockholm 2011, Sweden.

You might also be interested in the following publications if you are studying mobile application usage. Some of them are based on the AppSensor:

• Luis A. Leiva, Matthias Böhmer, Sven Gehring, Antonio Krüger. Back to the App: The Costs of Mobile Application Interruptions. In: Proceedings of MobileHCI ’12, San Francisco 2012, USA.
• Matthias Böhmer, Antonio Krüger. A Study on Icon Arrangement by Smartphone Users. In: Proceedings of CHI ’13. Paris 2013, France.
• Abhinav Parate, Matthias Böhmer, David Chu, Deepak Ganesan, Benjamin Marlin. Practical Prediction and Prefetch for Faster Access to Applications on Mobile phones. In: Proceedings of UbiComp ’13. Zürich 2013, Switzerland.
• Matthias Böhmer, Lyubomir Ganev, Antonio Krüger. AppFunnel: A Framework for Usage-centric Evaluation of Recommender Systems that Suggest Mobile Applications. In: Proceedings of IUI ’13. Santa Clara 2013, USA.

You might also want to look into the AWARE FRAMEWORK or the funf plattform.