Data Science

The mobile ecosystem is dramatically growing towards an unprecedented scale, with an extremely crowded market and fierce competition among app developers. Today, keeping users engaged with a mobile app is key for its success since users can remain active consumers of services and/or producers of new contents. However, users may abandon a mobile app at any time due to various reasons, e.g., the success of competing apps, decrease of interest in the provided services, etc. In this context, predicting when a user may get disengaged from an app is an invaluable resource for developers, creating the opportunity to apply intervention strategies aiming at recovering from disengagement (e.g., sending push notifications with new contents).In this study, we aim at providing evidence that predicting when mobile app users get disengaged is possible with a good level of accuracy. Specifically, we propose, apply, and evaluate a framework to model and predict User Engagement (UE) in mobile applications via different numerical models. The proposed framework is composed of an optimized agglomerative hierarchical clustering model coupled to (i) a Cox proportional hazards, (ii) a negative binomial, (iii) a random forest, and (iv) a boosted-tree model. The proposed framework is empirically validated by means of a year-long observational dataset collected from a real deployment of a waste recycling app. Our results show that in this context the optimized clustering model classifies users adequately and improves UE predictability for all numerical models. Also, the highest levels of prediction accuracy and robustness are obtained by applying either the random forest classifier or the boosted-tree algorithm.

New, multi-channel personal data sources (like heart rate, sleep patterns, travel patterns, or social activities) are enabled by ever increased availability of miniaturised technologies embedded within smartphones and wearables. These data sources enable personal self-management of lifestyle choices (e.g., exercise, move to a bike-friendly area) and, on a large scale, scientific discoveries to improve health and quality of life. However, there are no simple and reliable ways for individuals to securely collect, explore and share these sources. Additionally, much data is also wasted, especially when the technology provider ceases to exist, leaving the users without any opportunity to retrieve own datasets from “dead” devices or systems. Our research reveals evidence of what we term human data bleeding and offers guidance on how to address current issues by reasoning upon five core aspects, namely technological, financial, legal, institutional and cultural factors. To this end, we present preliminary specifications of an open platform for personal data storage and quality of life research. The Open Health Archive (OHA) is a platform that would support individual, community and societal needs by facilitating collecting, exploring and sharing personal health and QoL data.

The major application areas of reinforcement learning (RL) have traditionally been game playing and continuous control. In recent years, however, RL has been increasingly applied in systems that interact with humans. RL can personalize digital systems to make them more relevant to individual users. Challenges in personalization settings may be different from challenges found in traditional application areas of RL. An overview of work that uses RL for personalization, however, is lacking. In this work, we introduce a framework of personalization settings and use it in a systematic literature review. Besides setting, we review solutions and evaluation strategies. Results show that RL has been increasingly applied to personalization problems and realistic evaluations have become more prevalent. RL has become sufficiently robust to apply in contexts that involve humans and the field as a whole is growing. However, it seems not to be maturing: the ratios of studies that include a comparison or a realistic evaluation are not showing upward trends and the vast majority of algorithms are used only once. This review can be used to find related work across domains, provides insights into the state of the field and identifies opportunities for future work.