Alhajj, RedaSahin, Coskun2021-09-142021-09-142021-09-07Sahin, C. (2021). Social Media Emergency Analysis and Realistic Evacuation Modeling (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.http://hdl.handle.net/1880/113864The widespread of disasters necessitates appropriate actions to be taken to avoid casualties or at least reduce them to the minimum level possible. In this work, we propose solutions to two of the ways we can help it. The first part is using the potential of social media to detect emergencies and provide further information for the public and rescue teams. It uses a multi-layer machine learning approach to find and cluster emergency-related messages. Natural language processing and information extraction techniques are adopted for location detection, casualty and severity calculation. The effectiveness of the model is shown on Twitter data in nearly real-time using its own API. The second part of the thesis investigates the area of crowd behavior modeling in order to provide a platform for simulating various emergency evacuation scenarios. Crowd behavior modeling is an important challenge especially for games, social simulation and military training software. There are various applications focusing on specific approaches to solve social, physical, emotional and cognitive dimensions of this problem. In this work, first, we built and agent-based framework that adopts OCC emotion model and Belief-Desire-Intention (BDI) approach in a heterogeneous environment containing individuals with different knowledge of the surroundings. It uses a deep Q-learning model running on top of a neural network for creating a set of partially-trained agents. Second, we simulate group decision-making process using a mathematical ferromagnetism model and analyze how it affects the overall success of the crowd achieving a goal together. Our work contributes to the literature in two different ways. Firstly, it shows that a multi-layer classification model is effective for detecting emergencies in real-time using unstructured data. Moreover, it provides evacuation simulation scenarios on the public buildings where an emergency occurs and saves valuable time for rescue teams while planning an emergency response. Secondly, it combines state-of-the-art frameworks for crowd behavior modeling with a partial learning mechanism. Our experiments show that the system is capable of simulating common group patterns during emergencies. Moreover, it provides a generic crowd framework, where partially-trained agents can find optimal solutions via interaction and collaboration.engUniversity of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission.Machine learningSocial media analysisInformation extractionEmergency detectionCrowd behavior modelingMulti-agent systemsEvacuation simulationDeep Q-learningComputer Sciencedoctoral thesis10.11575/PRISM/39190