The surge in the application of big data and predictive analytics in fields of crisis management, such as pandemics and epidemics, highlights the vital need for advanced research in these areas, particularly in the wake of the COVID-19 pandemic. Traditional methods, which typically rely on historical data to forecast future trends, fall short in addressing the complex and ever-changing nature of challenges like pandemics and public health crises. This inadequacy is further underscored by the pandemic's significant impact on various sectors, notably healthcare, government, and the hotel industry. Current models often overlook key factors such as static spatial elements, socioeconomic conditions, and the wealth of data available from social media, which are crucial for a comprehensive understanding and effective response to these multifaceted crises. This thesis employs spatial forecasting and predictive analytics to address crisis management in several distinct but interrelated contexts: the COVID-19 pandemic, the opioid crisis, and the impact of the pandemic on the hotel industry. The first part of the study focuses on using big data analytics to explore the relationship between socioeconomic factors and the spread of COVID-19 at the zip code level, aiming to predict high-risk areas for infection. The second part delves into the opioid crisis, utilizing semi-supervised deep learning techniques to monitor and categorize drug-related discussions on Reddit. The third part concentrates on developing spatial forecasting and providing explanations of the rising epidemic of drug overdose fatalities. The fourth part of the study extends to the realm of the hotel industry, aiming to optimize customer experience by analyzing online reviews and employing a localized Large Language Model to generate future customer trends and scenarios. Across these studies, the thesis aims to provide actionable insights and comprehensive solutions for effectively managing these major crises. For the first work, the majority of current research in pandemic modeling primarily relies on historical data to predict dynamic trends such as COVID-19. This work makes the following contributions in spatial COVID-19 pandemic forecasting: 1) the development of a unique model solely employing a wide range of socioeconomic indicators to forecast areas most susceptible to COVID-19, using detailed static spatial analysis, 2) identification of the most and least influential socioeconomic variables affecting COVID-19 transmission within communities, 3) construction of a comprehensive dataset that merges state-level COVID-19 statistics with corresponding socioeconomic attributes, organized by zip code. For the second work, we make the following contributions in detecting drug Abuse crisis via social media: 1) enhancing the Dynamic Query Expansion (DQE) algorithm to dynamically detect and extract evolving drug names in Reddit comments, utilizing a list curated from government and healthcare agencies, 2) constructing a textual Graph Convolutional Network combined with word embeddings to achieve fine-grained drug abuse classification in Reddit comments, identifying seven specific drug classes for the first time, 3) conducting extensive experiments to validate the framework, outperforming six baseline models in drug abuse classification and demonstrating effectiveness across multiple types of embeddings. The third study focuses on developing spatial forecasting and providing explanations of the escalating epidemic of drug overdose fatalities. Current research in this field has shown a deficiency in comprehensive explanations of the crisis, spatial analyses, and predictions of high-risk zones for drug overdoses. Addressing these gaps, this study contributes in several key areas: 1) Establishing a framework for spatially forecasting drug overdose fatalities predominantly affecting U.S. counties, 2) Proposing solutions for dealing with scarce and heterogeneous data sets, 3) Developing an algorithm that offers clear and actionable insights into the crisis, and 4) Conducting extensive experiments to validate the effectiveness of our proposed framework. In the fourth study, we address the profound impact of the pandemic on the hotel industry, focusing on the optimization of customer experience. Traditional methodologies in this realm have predominantly relied on survey data and limited segments of social media analytics. Those methods are informative but fall short of providing a full picture due to their inability to include diverse perspectives and broader customer feedback. Our study aims to make the following contributions: 1) the development of an integrated platform that distinguishes and extracts positive and negative Memorable Experiences (MEs) from online customer reviews within the hotel industry, 2) The incorporation of an advanced analytical module that performs temporal trend analysis of MEs, utilizing sophisticated data mining algorithms to dissect customer feedback on a monthly and yearly scale, 3) the implementation of an advanced tool that generates prospective and unexplored Memorable Experiences (MEs) by utilizing a localized Large Language Model (LLM) with keywords extracted from authentic customer experiences to aid hotel management in preparing for future customer trends and scenarios. Building on the integrated predictive modeling approaches developed in the earlier parts of this dissertation, this final section explores the significant impacts of the COVID-19 pandemic on the airline industry. The pandemic has precipitated substantial financial losses and operational disruptions, necessitating innovative crisis management strategies within this sector. This study introduces a novel analytical framework, EAGLE (Enhancing Airline Groundtruth Labels and Review rating prediction), which utilizes Large Language Models (LLMs) to improve the accuracy and objectivity of customer sentiment analysis in strategic airline route planning. EAGLE leverages LLMs for zero-shot pseudo-labeling and zero-shot text classification, to enhance the processing of customer reviews without the biases of manual labeling. This approach streamlines data analysis, and refines decision-making processes which allows airlines to align route expansions with nuanced customer preferences and sentiments effectively. The comprehensive application of LLMs in this context underscores the potential of predictive analytics to transform traditional crisis management strategies by providing deeper, more actionable insights. / Doctor of Philosophy / In today's digital age, where vast amounts of data are generated every second, understanding and managing crises like pandemics or economic disruptions has become increasingly crucial. This dissertation explores the use of advanced predictive modeling and analytics to manage various crises, significantly enhancing how predictions and responses to these challenges are developed. The first part of the research uses data analysis to identify areas at higher risk during the COVID-19 pandemic, focusing on how different socioeconomic factors can affect virus spread at a local level. This approach moves beyond traditional methods that rely on past data, providing a more dynamic way to forecast and manage public health crises. The study then examines the opioid crisis by analyzing social media platforms like Reddit. Here, a method was developed to automatically detect and categorize discussions about drug abuse. This technique aids in understanding how drug-related conversations evolve online, providing insights that could guide public health responses and policy-making. In the hospitality sector, customer reviews were analyzed to improve service quality in hotels. By using advanced data analysis tools, key trends in customer experiences were identified, which can help businesses adapt and refine their services in real-time, enhancing guest satisfaction. Finally, the study extends to the airline industry, where a model was developed that uses customer feedback to improve airline services and route planning. This part of the research shows how sophisticated analytics can help airlines better understand and meet traveler needs, especially during disruptions like the pandemic. Overall, the dissertation provides methods to better manage crises and illustrates the vast potential of predictive analytics in making informed decisions that can significantly mitigate the impacts of future crises. This research is vital for anyone—from government officials to business leaders—looking to harness the power of data for crisis management and decision-making.

Les médias sociaux ont largement modifié la manière dont nous produisons, diffusons et consommons l'information et sont de fait devenus des vecteurs d'information importants. L’objectif de cette thèse est d’aider à la compréhension du phénomène de diffusion de l’information dans les médias sociaux, en fournissant des moyens d’analyse et de modélisation.Premièrement, nous proposons MABED, une méthode statistique pour détecter automatiquement les évènements importants qui suscitent l'intérêt des utilisateurs des médias sociaux à partir du flux de messages qu'ils publient, dont l'originalité est d'exploiter la fréquence des interactions sociales entre utilisateurs, en plus du contenu textuel des messages. Cette méthode diffère par ailleurs de celles existantes en ce qu'elle estime dynamiquement la durée de chaque évènement, plutôt que de supposer une durée commune et fixée à l'avance pour tous les évènements. Deuxièmement, nous proposons T-BASIC, un modèle probabiliste basé sur la structure de réseau sous-jacente aux médias sociaux pour prédire la diffusion de l'information, plus précisément l'évolution du volume d'utilisateurs relayant une information donnée au fil du temps. Contrairement aux modèles similaires également basés sur la structure du réseau, la probabilité qu'une information donnée se diffuse entre deux utilisateurs n'est pas constante mais dépendante du temps. Nous décrivons aussi une procédure pour l'inférence des paramètres latents du modèle, dont l'originalité est de formuler les paramètres comme des fonctions de caractéristiques observables des utilisateurs. Troisièmement, nous proposons SONDY, un logiciel libre et extensible implémentant des méthodes tirées de la littérature pour la fouille et l'analyse des données issues des médias sociaux. Le logiciel manipule deux types de données : les messages publiés par les utilisateurs, et la structure du réseau social interconnectant ces derniers. Contrairement aux logiciels académiques existants qui se concentrent soit sur l'analyse des messages, soit sur l'analyse du réseau, SONDY permet d'analyser ces deux types de données conjointement en permettant l'analyse de l'influence par rapport aux évènements détectés. Les expérimentations menées à l'aide de divers jeux de données collectés sur le média social Twitter démontrent la pertinence de nos propositions et mettent en lumière des propriétés qui nous aident à mieux comprendre les mécanismes régissant la diffusion de l'information. Premièrement, en comparant les performances de MABED avec celles de méthodes récentes tirées de la littérature, nous montrons que la prise en compte des interactions sociales entre utilisateurs conduit à une détection plus précise des évènements importants, avec une robustesse accrue en présence de contenu bruité. Nous montrons également que MABED facilite l'interprétation des évènements détectés en fournissant des descriptions claires et précises, tant sur le plan sémantique que temporel. Deuxièmement, nous montrons la validité de la procédure proposée pour estimer les probabilités de diffusion sur lesquelles repose le modèle T-BASIC, en illustrant le pouvoir prédictif des caractéristiques des utilisateurs sélectionnées et en comparant les performances de la méthode d'estimation proposée avec celles de méthodes tirées de la littérature. Nous montrons aussi l'intérêt d'avoir des probabilités non constantes, ce qui permet de prendre en compte dans T-BASIC la fluctuation du niveau de réceptivité des utilisateurs des médias sociaux au fil du temps. Enfin, nous montrons comment, et dans quelle mesure, les caractéristiques sociales, thématiques et temporelles des utilisateurs affectent la diffusion de l'information. Troisièmement, nous illustrons à l'aide de divers scénarios l'utilité du logiciel SONDY, autant pour des non-experts, grâce à son interface utilisateur avancée et des visualisations adaptées, que pour des chercheurs du domaine, grâce à son interface de programmation. / Social media have greatly modified the way we produce, diffuse and consume information, and have become powerful information vectors. The goal of this thesis is to help in the understanding of the information diffusion phenomenon in social media by providing means of modeling and analysis.First, we propose MABED (Mention-Anomaly-Based Event Detection), a statistical method for automatically detecting events that most interest social media users from the stream of messages they publish. In contrast with existing methods, it doesn't only focus on the textual content of messages but also leverages the frequency of social interactions that occur between users. MABED also differs from the literature in that it dynamically estimates the period of time during which each event is discussed rather than assuming a predefined fixed duration for all events. Secondly, we propose T-BASIC (Time-Based ASynchronous Independent Cascades), a probabilistic model based on the network structure underlying social media for predicting information diffusion, more specifically the evolution of the number of users that relay a given piece of information through time. In contrast with similar models that are also based on the network structure, the probability that a piece of information propagate from one user to another isn't fixed but depends on time. We also describe a procedure for inferring the latent parameters of that model, which we formulate as functions of observable characteristics of social media users. Thirdly, we propose SONDY (SOcial Network DYnamics), a free and extensible software that implements state-of-the-art methods for mining data generated by social media, i.e. the messages published by users and the structure of the social network that interconnects them. As opposed to existing academic tools that either focus on analyzing messages or analyzing the network, SONDY permits the joint analysis of these two types of data through the analysis of influence with respect to each detected event.The experiments, conducted on data collected on Twitter, demonstrate the relevance of our proposals and shed light on some properties that give us a better understanding of the mechanisms underlying information diffusion. First, we compare the performance of MABED against those of methods from the literature and find that taking into account the frequency of social interactions between users leads to more accurate event detection and improved robustness in presence of noisy content. We also show that MABED helps with the interpretation of detected events by providing clearer textual description and more precise temporal descriptions. Secondly, we demonstrate the relevancy of the procedure we propose for estimating the pairwise diffusion probabilities on which T-BASIC relies. For that, we illustrate the predictive power of users' characteristics, and compare the performance of the method we propose to estimate the diffusion probabilities against those of state-of-the-art methods. We show the importance of having non-constant diffusion probabilities, which allows incorporating the variation of users' level of receptivity through time into T-BASIC. We also study how -- and in which proportion -- the social, topical and temporal characteristics of users impact information diffusion. Thirdly, we illustrate with various scenarios the usefulness of SONDY, both for non-experts -- thanks to its advanced user interface and adapted visualizations -- and for researchers -- thanks to its application programming interface.

This dissertation thesis deals with the application of the concept of information distance to the social network data analysis. We consider this data as recorded acts of social action. As such, they express certain attitudes, values and intentions. We introduce a formula for calculating the Normalized Social Distance and, based on the series of case studies, we prove the usefulness and validity of this approach. The application of formal mathematical and computer science techniques to massive data records of human action in social network environments is enabled by the change brought by new media and the associated technological advancement. This change is accompanied by a gradual transition of research methods in the humanities, referred to as the onset of digital humanities. This approach is characterized by the application of quantitative methods in the field of humanities and the discovery of new data areas useful for analyses. In case of social media data, the differentiation between quantitative and qualitative methods is no longer valid. A good example is also this thesis, in which information theory specifically combines the methods of a traditional social network analysis and the Goffman's frame analysis of human action. Keywords Information distance, Normalized Social Distance, Kolmogorov...