Assessing the impact of mobility on epidemic spreading is of crucial importance for understanding the effect of policies like mass quarantines and selective re-openings. While many factors affect disease incidence at a local level, making it more or less homogeneous with respect to other areas, the importance of multi-seeding has often been overlooked. Multi-seeding occurs when several independent (non-clustered) infected individuals arrive at a susceptible population. This can lead to independent outbreaks that spark from distinct areas of the local contact (social) network. Such mechanism has the potential to boost incidence, making control efforts and contact tracing less effective. Here, through a modeling approach we show that the effect produced by the number of initial infections is non-linear on the incidence peak and peak time. When case importations are carried by mobility from an already infected area, this effect is further enhanced by the local demography and underlying mixing patterns: the impact of every seed is larger in smaller populations. Finally, both in the model simulations and the analysis, we show that a multi-seeding effect combined with mobility restrictions can explain the observed spatial heterogeneities in the first wave of COVID-19 incidence and mortality in five European countries. Our results allow us for identifying what we have called epidemic epicenter: an area that shapes incidence and mortality peaks in the entire country. The present work further clarifies the nonlinear effects that mobility can have on the evolution of an epidemic and highlight their relevance for epidemic control.

Data are building blocks to information and, subsequently, they are vital input to knowledge. Today, in the midst of the digital era, vast quantities of highly-complex data are being collected and processed at an unprecedented scale. This abundance of data has highlighted the importance of efficient and effective knowledge-discovery algorithms to identify patterns hidden in the data with the ultimate aim of uncovering valuable knowledge and shape our understanding of the world around us. To capitalize on the opportunities offered by massive amounts of data as well as modern computing power, for many years, research in knowledge discovery and related areas has introduced algorithms that are increasingly efficient and effective, but also more and more opaque and unpredictable. Recently, growing interest in the ethical dimensions of algorithms has drawn attention to the limitations of opaque algorithms and has emphasized a need for trustworthy algorithms particularly when such algorithms are used to support high-stakes decision making. In order to be trustworthy, algorithms should solve a clearly defined problem via a clear sequence of instructions, they should not be utterly unsuccessful in any particular case and they should be easy to understand and interpret for humans so that no harmful biases can be hidden. In this thesis, we pursue the goal of developing novel knowledge-discovery algorithmic methods that are not only highly efficient to face the challenges and opportunities posed by modern data, but also trustworthy. In particular, we propose efficient and trustworthy methods for a collection of popular knowledgediscovery tasks. First, we consider tasks of exact inference in Bayesian networks and hidden Markov models. Trustworthy approaches for such tasks exist. However, their applicability may be severely limited by time or memory requirements. Therefore, we propose novel methods to reduce the time or memory resources that are needed by existing approaches for the considered exact inference tasks. Beside exact inference tasks, we also consider two different knowledge-discovery tasks that arise naturally in modern data: multi-label classification and community search in temporal graphs. Regarding multi-label classification, we propose an efficient and accurate rule-based multi-label classifier that drastically improves upon the interpretability of existing solutions. For community search in temporal graphs, we formalise the task for the first time, and we propose a solution that guarantees high efficiency and interpretability. In designing knowledge-discovery methods, we often rely on existing database-management and probabilistic methods. Methods for database management are valuable to address the large dimension and high complexity of modern data, while probabilistic methods are essential to methodologically handle uncertainty in the data.

In recent years, many studies have drawn attention to the important role of collective awareness and human behaviour during epidemic outbreaks. A number of modelling efforts have investigated the interaction between the disease transmission dynamics and human behaviour change mediated by news coverage and by information spreading in the population. Yet, given the scarcity of data on public awareness during an epidemic, few studies have relied on empirical data. Here, we use fine-grained, geo-referenced data from three online sources-Wikipedia, the GDELT Project and the Internet Archive-to quantify population-scale information seeking about the 2016 Zika virus epidemic in the U.S., explicitly linking such behavioural signal to epidemiological data. Geo-localized Wikipedia pageview data reveal that visiting patterns of Zika-related pages in Wikipedia were highly synchronized across the United States and largely explained by exposure to national television broadcast. Contrary to the assumption of some theoretical epidemic models, news volume and Wikipedia visiting patterns were not significantly correlated with the magnitude or the extent of the epidemic. Attention to Zika, in terms of Zika-related Wikipedia pageviews, was high at the beginning of the outbreak, when public health agencies raised an international alert and triggered media coverage, but subsequently exhibited an activity profile that suggests nonlinear dependencies and memory effects in the relation between information seeking, media pressure, and disease dynamics. This calls for a new and more general modelling framework to describe the interaction between media exposure, public awareness and disease dynamics during epidemic outbreaks.

Seasonal influenza surveillance is usually carried out by sentinel general practitioners (GPs) who compile weekly reports based on the number of influenza-like illness (ILI) clinical cases observed among visited patients. This traditional practice for surveillance generally presents several issues, such as a delay of one week or more in releasing reports, population biases in the health-seeking behaviour, and the lack of a common definition of ILI case. On the other hand, the availability of novel data streams has recently led to the emergence of non-traditional approaches for disease surveillance that can alleviate these issues. In Europe, a participatory web-based surveillance system called Influenzanet represents a powerful tool for monitoring seasonal influenza epidemics thanks to aid of self-selected volunteers from the general population who monitor and report their health status through Internet-based surveys, thus allowing a real-time estimate of the level of influenza circulating in the population. In this work, we propose an unsupervised probabilistic framework that combines time series analysis of self-reported symptoms collected by the Influenzanet platforms and performs an algorithmic detection of groups of symptoms, called syndromes. The aim of this study is to show that participatory web-based surveillance systems are capable of detecting the temporal trends of influenza-like illness even without relying on a specific case definition. The methodology was applied to data collected by Influenzanet platforms over the course of six influenza seasons, from 2011-2012 to 2016-2017, with an average of 34,000 participants per season. Results show that our framework is capable of selecting temporal trends of syndromes that closely follow the ILI incidence rates reported by the traditional surveillance systems in the various countries (Pearson correlations ranging from 0.69 for Italy to 0.88 for the Netherlands, with the sole exception of Ireland with a correlation of 0.38). The proposed framework was able to forecast quite accurately the ILI trend of the forthcoming influenza season (2016-2017) based only on the available information of the previous years (2011-2016). Furthermore, to broaden the scope of our approach, we applied it both in a forecasting fashion to predict the ILI trend of the 2016-2017 influenza season (Pearson correlations ranging from 0.60 for Ireland and UK, and 0.85 for the Netherlands) and also to detect gastrointestinal syndrome in France (Pearson correlation of 0.66). The final result is a near-real-time flexible surveillance framework not constrained by any specific case definition and capable of capturing the heterogeneity in symptoms circulation during influenza epidemics in the various European countries.

When analyzing temporal networks, a fundamental task is the identification of dense structures (i.e., groups of vertices that exhibit a large number of links), together with their temporal span (i.e., the period of time for which the high density holds). We tackle this task by introducing a notion of temporal core decomposition where each core is associated with its span: we call such cores span-cores. As the total number of time intervals is quadratic in the size of the temporal domain T under analysis, the total number of span-cores is quadratic in $|T|$ as well. Our first contribution is an algorithm that, by exploiting containment properties among span-cores, computes all the span-cores efficiently. Then, we focus on the problem of finding only the maximal span-cores, i.e., span-cores that are not dominated by any other span-core by both the coreness property and the span. We devise a very efficient algorithm that exploits theoretical findings on the maximality condition to directly compute the maximal ones without computing all span-cores. Experimentation on several real-world temporal networks confirms the efficiency and scalability of our methods. Applications on temporal networks, gathered by a proximity-sensing infrastructure recording face-to-face interactions in schools, highlight the relevance of the notion of (maximal) span-core in analyzing social dynamics and detecting/correcting anomalies in the data.

Today, many people spend a lot of time online. Their social interactions captured in online social networks are an important part of the overall personal social profile, in addition to interactions taking place offline. This paper investigates whether relations captured by online social networks can be used as a proxy for the relations in offline social networks, such as networks of human face-to-face (F2F) proximity and coauthorship networks. Particularly, the paper focuses on interactions of computer scientists in online settings (homepages, social networks profiles and connections) and offline settings (scientific collaboration, face-to-face communications during the conferences). We focus on quantitative studies and investigate the structural similarities and correlations of the induced networks; in addition, we analyze implications between networks. Finally, we provide a qualitative user analysis to find characteristics of good and bad proxies.

Micro-blogging systems such as Twitter expose digital traces of social discourse with an unprecedented degree of resolution of individual behaviors. They offer an opportunity to investigate how a large-scale social system responds to exogenous or endogenous stimuli, and to disentangle the temporal, spatial and topical aspects of users' activity. Here we focus on spikes of collective attention in Twitter, and specifically on peaks in the popularity of hashtags. Users employ hashtags as a form of social annotation, to define a shared context for a specific event, topic, or meme. We analyze a large-scale record of Twitter activity and find that the evolution of hashtag popularity over time defines discrete classes of hashtags. We link these dynamical classes to the events the hashtags represent and use text mining techniques to provide a semantic characterization of the hashtag classes. Moreover, we track the propagation of hashtags in the Twitter social network and find that epidemic spreading plays a minor role in hashtag popularity, which is mostly driven by exogenous factors.

Mobile, social, real-time: the ongoing revolution in the way people communicate has given rise to a new kind of epidemiology. Digital data sources, when harnessed appropriately, can provide local and timely information about disease and health dynamics in populations around the world. The rapid, unprecedented increase in the availability of relevant data from various digital sources creates considerable technical and computational challenges.

Social media have attracted considerable attention because their open-ended nature allows users to create lightweight semantic scaffolding to organize and share content. To date, the interplay of the social and topical components of social media has been only partially explored. Here, we study the presence of homophily in three systems that combine tagging social media with online social networks. We find a substantial level of topical similarity among users who are close to each other in the social network. We introduce a null model that preserves user activity while removing local correlations, allowing us to disentangle the actual local similarity between users from statistical effects due to the assortative mixing of user activity and centrality in the social network. This analysis suggests that users with similar interests are more likely to be friends, and therefore topical similarity measures among users based solely on their annotation metadata should be predictive of social links. We test this hypothesis on several datasets, confirming that social networks constructed from topical similarity capture actual friendship accurately. When combined with topological features, topical similarity achieves a link prediction accuracy of about 92%.

Large medical conferences offer opportunities for participants to find industry exhibitors that offer products and services relevant to their professional interests. Companies often invest significant effort in promotions that encourage participants to spend time at their stand (e.g. providing free gifts, leaflets, running competitions) and register some contact details. Attendees will use the conference to find others who also share similar professional interests, as well as keep up to date with developments on products such has pharmaceuticals and medical equipment. From both perspectives, a number of improvements can be made to enhance the overall experience by using existing active RFID technology: Vendors would be able to more closely monitor the success of their promotions with statistics on the stand's visitors, as well as find more potential customers by using real-time visualizations; Participants would be able to log their social interactions, keeping an electronic history of the people they have met. The SocioPatterns project and Live Social Semantics experiments have recently demonstrated a scalable and robust infrastructure that would support these kinds of improvements. In this paper, we propose an infrastructure that provides enhanced social interaction services for vendors and participants by using small active RFID badges worn by attendees and attached to fixed locations.

Web 2.0 applications have attracted a considerable amount of attention because their open-ended nature allows users to create lightweight semantic scaffolding to organize and share content. To date, the interplay of the social and semantic components of social media has been only partially explored. Here we focus on Flickr and Last.fm, two social media systems in which we can relate the tagging activity of the users with an explicit representation of their social network. We show that a substantial level of local lexical and topical alignment is observable among users who lie close to each other in the social network. We introduce a null model that preserves user activity while removing local correlations, allowing us to disentangle the actual local alignment between users from statistical effects due to the assortative mixing of user activity and centrality in the social network. This analysis suggests that users with similar topical interests are more likely to be friends, and therefore semantic similarity measures among users based solely on their annotation metadata should be predictive of social links. We test this hypothesis on the Last.fm data set, confirming that the social network constructed from semantic similarity captures actual friendship more accurately than Last.fm's suggestions based on listening patterns.

Social interactions are one of the key factors to the success of conferences and similar community gatherings. This paper describes a novel application that integrates data from the semantic web, online social networks, and a real-world contact sensing platform. This application was successfully deployed at ESWC09, and actively used by 139 people. Personal profiles of the participants were automatically generated using several Web 2.0 systems and semantic academic data sources, and integrated in real-time with face-to-face contact networks derived from wearable sensors. Integration of all these heterogeneous data layers made it possible to offer various services to conference attendees to enhance their social experience such as visualisation of contact data, and a site to explore and connect with other participants. This paper describes the architecture of the application, the services we provided, and the results we achieved in this deployment. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Social bookmarking systems are becoming increasingly important data sources for bootstrapping and maintaining Semantic Web applications. Their emergent information structures have become known as folksonomies. A key question for harvesting semantics from these systems is how to extend and adapt traditional notions of similarity to folksonomies, and which measures are best suited for applications such as community detection, navigation support, semantic search, user profiling and ontology learning. Here we build an evaluation framework to compare various general folksonomy-based similarity measures, which are derived from several established information-theoretic, statistical, and practical measures. Our framework deals generally and symmetrically with users, tags, and resources. For evaluation purposes we focus on similarity between tags and between resources and consider different methods to aggregate annotations across users. After comparing the ability of several tag similarity measures to predict user-created tag relations, we provide an external grounding by user-validated semantic proxies based on WordNet and the Open Directory Project. We also investigate the issue of scalability. We find that mutual information with distributional micro-aggregation across users yields the highest accuracy, but is not scalable; per-user projection with collaborative aggregation provides the best scalable approach via incremental computations. The results are consistent across resource and tag similarity.