Protecting Mobile Food Diaries from Getting too Personal

Abstract

Smartphone applications that use passive sensing to support human health and well-being primarily rely on: (a) generating lowdimensional representations from high-dimensional data streams; (b) making inferences regarding user behavior; and (c) using those inferences to benefit application users. Meanwhile, sometimes these datasets are shared with third parties as well. Human-centered ubiquitous systems need to ensure that sensitive attributes of users are protected when applications provide utility to people based on such behavioral inferences. In this paper, we demonstrate that inferences of sensitive attributes of users (gender, body mass index category) are possible using low-dimensional and sparse data coming from mobile food diaries (a combination of sensor data and self-reports). After exposing this potential risk, we demonstrate how deep learning techniques can be used for feature transformation to preserve sensitive user information while achieving high accuracies for application-related inferences (e.g. inferring the type of consumed food). Our work is based on two datasets of daily eating behavior of 160 young adults from Switzerland (N =122) and Mexico (N =38). Results show that using the proposed approach, accuracies in the order of 75%-90% can be achieved for application related inferences, while reducing the sensitive inference to almost random performance. CCS CONCEPTS• Human-centered computing → Mobile computing; Smartphones; Mobile phones; Empirical studies in ubiquitous and mobile computing; • Social and professional topics → Gender; • Applied computing → Consumer health; Health informatics; Sociology.

Alone or With Others? Understanding Eating Episodes of College Students with Mobile Sensing

“…Furthermore, automatically inferring attributes related to social context would enable mobile food diaries to send contextaware notifications [21] and to support interventions [30], and also to help users adhere to healthy eating practices [14,20]. In this study, similar to prior mHealth sensing studies with food diaries [5,25,33], we consider eating to be a holistic event, and use a binary categorization for the social context of eatingeating-alone vs. eating-with-others as a construct to understand food consumption behavior of college students in two countries. Hence, this paper has two contributions: Contribution 1: We conducted a data analysis of wearable and mobile sensing datasets collected from 206 college students of two arXiv:2011.11694v2 [cs.CY] 28 Nov 2020 countries (Switzerland and Mexico).…”

“…The average age of study participants was 23.4 years, and the cohort had 44% men and 56% women. A more detailed feature summary with naming conventions is available in [25].…”

“…In the CH-Dataset, activity level features captured using the wearable show significant mean differences for all four features shown here. For example, the distributions of feature mean steps bef (see Section 2 and [25] for feature descriptions) are different in terms of the shape, where eating-alone distribution is highly skewed towards lower values, meaning that activity levels are low for eating-alone episodes. A similar pattern can be seen for features such as min lightly bef and sd steps bef.…”

The Theory, Practice, and Ethical Challenges of Designing a Diversity-Aware Platform for Social Relations

“…Collecting smartphone data requires insights into local practices, as the use of smartphones among young people differ across countries (Mathur et al 2017;. Students may use distinct apps in some countries, or use mobile services differently because of the cost of phone devices and data plans, popular local trends, and culture at large (Meegahapola, Ruiz-Correa, and Gatica-Perez 2020b;2020a). Data collection needs to take these differences across cultures and countries into account.…”

Examining the Social Context of Alcohol Drinking in Young Adults with Smartphone Sensing

“…social context, concurrent activities, ambiance, location, etc.). For domains such as eating behavior, there are studies regarding both event detection (identifying eating events [9,77], inferring meal or snack episodes [12], inferring food categories [73]) and event characterization (inferring the social context around eating events [72]). Inferring mood [62,96] as well as identifying contexts around specific moods [24] has been attempted in ubicomp.…”