Keynote: Context, big data, and digital prejudices

In pervasive computing research and literature, context has mostly been seen as an information source for applications that adapt their behavior according to the current situation of their user or their (often physical) environment. This adaptation could be the change of the user interface, the performance of actions (like sending messages or triggering actuators), or the change of used resources (like network bandwidth or processing power). To determine relevant situations, many heterogeneous data sources could be used, ranging from sensor data over mined patterns in files to explicit user input. Since most sensors are not perfect, context quality has to be considered. And since many context-aware applications are mobile, the set of data sources may change during runtime. According to the widely used definition by Anind Dey, context can be “any information that can be used to characterize the situation of an entity”. In the past years, we have seen a significant increase in the so-called “big data” domain, in research, technology, and industrial usage. The desire to analyze, gain knowledge and use more and more data it in new ways is rising in a way that resemble a gold rush. Data is the new oil. Beside applications like predictive maintenance of machines or optimization of industrial processes, a main target for big data analyses are humans — in their roles as travelers, current or potential clients, or application users. We could say that big data is “any information that can be used to characterize the situation of a user”, and relate these approaches to what have been done in context modelling and reasoning. This gets even clearer when these analyses leave the virtual world (e.g., client behavior in web shops) and enter the real world (e.g., client behavior in retail). In addition to the ambiguities of the analysis itself that only leads to predictions with a limited probability, sensor data quality becomes an issue: the sensor data might be inaccurate, outdated or conflicting with other observations or physical laws; in addition, sensor data processing algorithms like object classification or tracking might lead to ambiguous results. In this talk, we will shortly review these two domains and derive what could be learned for context-aware applications. A special focus will be given on quality of context on all semantic levels, and how the improper consideration of quality issues can lead to dangerous digital prejudices.