Big Data, Small Personas: How Algorithms Shape the Demographic Representation of Data-Driven User Segments.

Derived from the notion of algorithmic bias, it is possible that creating user segments such as personas from data results in over- or under-representing certain segments (FAIRNESS), does not properly represent the diversity of the user populations (DIVERSITY), or produces inconsistent results when hyperparameters are changed (CONSISTENCY). Collecting user data on 363M video views from a global news and media organization, we compare personas created from this data using different algorithms. Results indicate that the algorithms fall into two groups: those that generate personas with low diversity-high fairness and those that generate personas with high diversity-low fairness. The algorithms that rank high on diversity tend to rank low on fairness (Spearman's correlation: -0.83). The algorithm that best balances diversity, fairness, and consistency is Spectral Embedding. The results imply that the choice of algorithm is a crucial step in data-driven user segmentation, because the algorithm fundamentally impacts the demographic attributes of the generated personas and thus influences how decision makers view the user population. The results have implications for algorithmic bias in user segmentation and creating user segments that not only consider commercial segmentation criteria but also consider criteria derived from ethical discussions in the computing community.

An Empirical Study of Questionnaires for the Diagnosis of Pediatric Obstructive Sleep Apnea.

Pediatric Obstructive Sleep Apnea (OSA) is a chronic disorder characterized by the disruption in sleep due to involuntary and temporary cessation of breathing. Definitive diagnosis of OSA requires an intrusive and expensive approach based on polysomnography where the children spend a night in the hospital under the supervision of a sleep technician. The prevalence of OSA is increasing, making the traditional diagnostic approach prohibitively expensive. There has been increasing interest in designing inexpensive approaches to screen children such as the use of questionnaires. In this paper, we study the efficacy of five widely used and representative questionnaires on their ability to diagnose and stratify OSA. Our experiments show that the diagnostic ability of each of these questionnaires is insufficient for widespread clinical use. Using techniques from data mining, we identify the most informative questions and propose a new questionnaire. We show that machine learning models trained based on the answers to our questionnaire can stratify OSA with higher accuracy.