Embedded Files
Introduction: Personality affects various social behaviors of an individual, such as collaboration, group dynamics, and social relationships within the workplace. However, existing methods including both traditional and AI powered systems for assessing personality have shortcomings:
self-assessed methods are cumbersome due to repeated assessment and erroneous due to a self-report bias.
On the other hand, automatic, data-driven personality detection raises privacy concerns due to a need for excessive personal data.
Yet, automatic peronality detection methods are gaining popularity due to the ease of use, effectiveness, and scalability of these solutions. While previous work on automatic personality detection has been instrumental in terms of accuracy, not enough attention is given to understanding users' perceptions of these systems. This project address these concerns by investigating users' privacy concerns in sharing group-specific data and their perception of using an automated personality assessment (APA) system which collects data unobtrusively. In doing so, we present an unobtrusive system for detecting personality within the workplace that combines user’s online and offline behaviors.
Position scatter plot of a participant
APA system pipeline
Technical aspects:
Android app: Wearable API 2.0Feature extraction: pythonModel building: sklearn, imblearn
Solution: We take a mix-method (survey + interviews) approach in understanding users' privacy concern and their perception of using APA system. We first administer a survey (N=89) to understand people's workspace social behaviors pertaining to their online and offline communications. Next, to cover various situations that user’s behaviors would be expressed differently, we collect data from four different online and offline data streams: online messenger usage data, online web/app usage data, offline location data and offline movement data. While collecting data, we consider two factors: privacy and unobtrusiveness. We try to minimize privacy issues that could arise which would make users feel intrusive. We also discuss the effect of different levels of unobtrusiveness, as high obtrusiveness might change user’s behavior while a completely unobtrusive measure, to not make users be aware that their data is collected, can be unethical. From the collected data, we extract behaviors that are psychologically relevant and analyze their correlations and significance in indicating one’s personality while weighing between privacy and accuracy of prediction. Through this work, we explore the design space of an APA system by considering privacy issues and unobtrusiveness during data collection
Conclusions: The survey result suggests that a) people engage in social behaviors differently in their online and offline channels, b) data-sharing concerns reduces significantly when n user have more control, c) yet, peoples' acceptance of sharing data vary with respect to richness of the data i.e. people feel less comfortable sharing videos than text. Further, users' privacy concerns regarding using APA systems stem from (1) potential misuse or intuitive discomfort without a clear reason (PY1), and (2) rational thinking around current scope/purpose of data collection. These concerns can be addressed by showing them raw data and by taking preventive measures while designining such systems.
Google Sites
Report abuse