Putting human behavior predictability in context

Abstract Various studies have investigated the predictability of different aspects human behavior such as mobility patterns, social interactions, and shopping online behaviors. However, existing researches been often limited to a single or combination few behavioral dimensions, they adopted perspective an outside observer who is unaware motivations behind specific behaviors activities given individual. The key assumption this work that deliberated based on individual’s own perception situation s/he in, therefore it should also be studied under same perspective. Taking inspiration from works in ubiquitous context-aware computing, we investigate role played by four contextual dimensions (or modalities), namely time, location, activity being carried out, ties, individuals’ behaviors, using month collected mobile phone sensor readings self-reported annotations about these modalities more than two hundred study participants. Our analysis shows any target modality (e.g. location) becomes substantially predictable when information other (time, activity, ties) made available. Multi-modality turns out some sense fundamental, values like “shopping”) are nearly impossible guess correctly unless known. Subjectivity has substantial impact predictability. A location recognition experiment suggests subjective convey ties objective derived GPS measurements. We conclude paper analyzing how identified allow compute diversity personal behavior, where show individuals easily rarer, rather frequent, context annotations. These results offer support favor developing innovative computational models enriched characterization behavior.