Background: The COVID-19 pandemic led to unprecedented mitigation efforts that disrupted the daily lives of millions. Beyond the general health repercussions of the pandemic itself, these measures also present a challenge to the world’s mental health and health care systems. Considering that traditional survey methods are time-consuming and expensive, we need timely and proactive data sources to respond to the rapidly evolving effects of health policy on our population’s mental health. Many people in the United States now use social media platforms such as Twitter to express the most minute details of their daily lives and social relations. This behavior is expected to increase during the COVID-19 pandemic, rendering social media data a rich field to understand personal well-being. Objective: This study aims to answer three research questions: (1) What themes emerge from a corpus of US tweets about COVID-19? (2) To what extent did social media use increase during the onset of the COVID-19 pandemic? and (3) Does sentiment change in response to the COVID-19 pandemic? Methods: We analyzed 86,581,237 public domain English language US tweets collected from an open-access public repository in three steps. First, we characterized the evolution of hashtags over time using latent Dirichlet allocation (LDA) topic modeling. Second, we increased the granularity of this analysis by downloading Twitter timelines of a large cohort of individuals (n=354,738) in 20 major US cities to assess changes in social media use. Finally, using this timeline data, we examined collective shifts in public mood in relation to evolving pandemic news cycles by analyzing the average daily sentiment of all timeline tweets with the Valence Aware Dictionary and Sentiment Reasoner (VADER) tool. Results: LDA topics generated in the early months of the data set corresponded to major COVID-19–specific events. However, as state and municipal governments began issuing stay-at-home orders, latent themes shifted toward US-related lifestyle changes rather than global pandemic-related events. Social media volume also increased significantly, peaking during stay-at-home mandates. Finally, VADER sentiment analysis scores of user timelines were initially high and stable but decreased significantly, and continuously, by late March. Conclusions: Our findings underscore the negative effects of the pandemic on overall population sentiment. Increased use rates suggest that, for some, social media may be a coping mechanism to combat feelings of isolation related to long-term social distancing. However, in light of the documented negative effect of heavy social media use on mental health, social media may further exacerbate negative feelings in the long-term for many individuals. Thus, considering the overburdened US mental health care structure, these findings have important implications for ongoing mitigation efforts.

Federated learning (FL) is a privacy-preserving distributed learning approach that allows multiple parties to jointly build machine learning models without disclosing sensitive data. Although FL has solved the problem of collaboration without compromising privacy, it has a significant communication overhead due to the repetitive updating of models during training. Several studies have proposed communication-efficient FL approaches to address this issue, but adequate solutions are still lacking in cases where parties must deal with different data features, also referred to as vertical federated learning (VFL). In this paper, we propose a communication-efficient approach for VFL that compresses the local data of clients, and then aggregates the compressed data from all clients to build an ML model. Since local data are shared in compressed form, the privacy of these data is preserved. Experiments on publicly available benchmark datasets using our proposed method show that the final model obtained by aggregation of compressed data from clients outperforms the performance of the local models of the clients.

Significance: Can entire societies become more or less depressed over time? Here, we look for the historical traces of cognitive distortions, thinking patterns that are strongly associated with internalizing disorders such as depression and anxiety, in millions of books published over the course of the last two centuries in English, Spanish, and German. We find a pronounced “hockey stick” pattern: Over the past two decades the textual analogs of cognitive distortions surged well above historical levels, including those of World War I and II, after declining or stabilizing for most of the 20th century. Our results point to the possibility that recent socioeconomic changes, new technology, and social media are associated with a surge of cognitive distortions.

Federated learning (FL) has emerged as a promising distributed learning paradigm with an added advantage of data privacy. With the growing interest in collaboration among data owners, FL has gained significant attention from organizations. The idea of FL is to enable collaborating participants train machine learning (ML) models on decentralized data without breaching privacy. In simpler words, federated learning is the approach of “bringing the model to the data, instead of bringing the data to the model”. Federated learning, when applied to data which is partitioned vertically across participants, is able to build a complete ML model by combining local models trained only using the data with distinct features at the local sites. This architecture of FL is referred to as vertical federated learning (VFL), which differs from the conventional FL on horizontally partitioned data. As VFL is different from conventional FL, it comes with its own issues and challenges. Motivated by the comparatively less explored side of FL, this paper provides a comprehensive overview of existing methods and developments in VFL, covering various aspects such as communication, learning, privacy, and applications. We conclude by identifying gaps in the current literature and proposing potential future directions for research in VFL.

Human sleep/wake cycles follow a stable circadian rhythm associated with hormonal, emotional, and cognitive changes. Changes of this cycle are implicated in many mental health concerns. In fact, the bidirectional relation between major depressive disorder and sleep has been well-documented. Despite a clear link between sleep disturbances and subsequent disturbances in mood, it is difficult to determine from self-reported data which specific changes of the sleep/wake cycle play the most important role in this association. Here we observe marked changes of activity cycles in millions of twitter posts of 688 subjects who explicitly stated in unequivocal terms that they had received a (clinical) diagnosis of depression as compared to the activity cycles of a large control group (n = 8791). Rather than a phase-shift, as reported in other work, we find significant changes of activity levels in the evening and before dawn. Compared to the control group, depressed subjects were significantly more active from 7 PM to midnight and less active from 3 to 6 AM. Content analysis of tweets revealed a steady rise in rumination and emotional content from midnight to dawn among depressed individuals. These results suggest that diagnosis and treatment of depression may focus on modifying the timing of activity, reducing rumination, and decreasing social media use at specific hours of the day.