Abstract: Rapid impact assessment in the immediate aftermath of a natural disaster is essential to provide adequate information to international organisations, local authorities, and first responders. Social media can support emergency response with evidence-based content posted by citizens and organisations during ongoing events. In the paper, we propose TriggerCit: an early flood alerting tool with a multilanguage approach focused on timeliness and geolocation. The paper focuses on assessing the reliability of the approach as a triggering system, comparing it with alternative sources for alerts, and evaluating the quality and amount of complementary information gathered. Geolocated visual evidence extracted from Twitter by TriggerCit was analysed in two case studies on floods in Thailand and Nepal in 2021. The system respectively returned a large scale and a local scale alert, both in a timely manner and accompanied by a valid geographical description, while providing information complementary to existing disaster alert mechanisms.

Abstract: This paper summarizes the final year of the four-year Text REtrieval Conference Incident Streams track (TREC-IS), which has produced a large dataset comprising 136,263 annotated tweets, spanning 98 crisis events. Goals of this final year were twofold: 1) to add new categories for assessing messages, with a focus on characterizing the audience, author, and images associated with these messages, and 2) to enlarge the TREC-IS dataset with new events, with an emphasis of deeper pools for sampling. Beyond these two goals, TREC-IS has nearly doubled the number of annotated messages per event for the 26 crises introduced in 2021 and has released a new parallel dataset of 312,546 images associated with crisis content – with 7,297 tweets having annotations about their embedded images. Our analyses of this new crisis data yields new insights about the context of a tweet; e.g., messages intended for a local audience and those that contain images of weather forecasts and infographics have higher than average assessments of priority but are relatively rare. Tweets containing images, however, have higher perceived priorities than tweets without images. Moving to deeper pools, while tending to lower classification performance, also does not generally impact performance rankings or alter distributions of information-types. We end this paper with a discussion of these datasets, analyses, their implications, and how they contribute both new data and insights to the broader crisis informatics community.

Abstract: Social media like Twitter offer not only an unprecedented amount of user-generated content covering developing emergencies but also act as a collector of news produced by heterogeneous sources, including big and small media companies as well as public authorities. However, this volume, velocity, and variety of data constitute the main value and, at the same time, the key challenge to implement and automatic detection and tracking of independent emergency events from the real-time stream of tweets. Leveraging online clustering and considering both textual and geographical features, we propose, implement, and evaluate an algorithm to automatically detect emergency events applying a ‘glocal’ approach, i.e., offering a global coverage while detecting events at local (municipality level) scale.

Abstract: Social media serves as an important communication channel for people affected by crises, creating a data source for emergency responders wanting to improve situational awareness. In particular, social listening on Twitter has been widely used for real-time analysis of crisis-related messages. This approach, however, is often hindered by the small fraction of (hyper-)localized content and by the inability to explicitly ask affected populations about aspects with the most operational value. Here, we explore a new form of social media data collected through targeted poll ads on Facebook. Using geo-targeted ads during flood events in six countries, we show that it is possible to collect thousands of poll responses within hours of launching the ad campaign, and at a cost of a few (US dollar) cents per response. We believe that this flexible, fast, and affordable data collection can serve as a valuable complement to existing approaches.

Abstract: Social media generates large amounts of almost real-time data which has proven valuable in disaster response. Specially for providing information within the first 48 hours after a disaster occurs. However, this potential is poorly exploited in operational environments due to the challenges of curating social media data. This work builds on top of the latest research on automatic classification of social media content, proposing the use of ensemble learning to help in the classification of social media images for disaster response. Ensemble methods use multiple learning algorithms to obtain better predictive performance than could be obtained from any of the constituent learning algorithms alone. Experimental results show that ensemble learning is a valuable technology for the analysis of social media images for disaster response,and could potentially ease the integration of social media data within an operational environment.

Abstract: Overload reduction is essential to exploit Twitter text data for crisis management. Often used pre-trained machine learning models require training data for both, crisis-related and off-topic content. However, this task can also be formulated as a one-class classification problem in which labeled off-topic samples are not required. Gaussian processes (GPs) have great potential in both, binary and one-class settings and are therefore investigated in this work. Deep kernel learning combines the representative power of text embeddings with the Bayesian formalism of GPs. Motivated by this, we investigate the potential of deep kernel models for the task of classifying crisis-related tweet texts with special emphasis on cross-event applications. Compared to standard binary neural networks, first experiments with one-class GP models reveal a great potential for realistic scenarios, offering a fast and flexible approach for interactive model training without requiring off-topic training samples and comprehensive expert knowledge (only two model parameters involved).

Abstract: Social media platforms contain abundant data that can provide comprehensive knowledge of historical and real-time events. During crisis events, the use of social media peaks, as people discuss what they have seen, heard, or felt. Previous studies confirm the usefulness of such socially generated discussions for the public, first responders, and decision-makers to gain a better understanding of events as they unfold at the ground level. This study performs an extensive analysis of COVID-19-related Twitter discussions generated in Australia between January 2020, and October 2022. We explore the Australian Twitterverse by employing state-of-the-art approaches from both supervised and unsupervised domains to perform network analysis, topic modeling, sentiment analysis, and causality analysis. As the presented results provide a comprehensive understanding of the Australian Twitterverse during the COVID-19 pandemic, this study aims to explore the discussion dynamics to aid the development of future automated information systems for epidemic/pandemic management.

Abstract: Between 2018 and 2021, the Incident Streams track (TREC-IS) developed standard approaches for classifying information types and criticality of tweets during crises. While successful in producing substantial collections of labeled data, TREC-IS as a data challenge had several limitations: It only evaluated information at type-level rather than what was reported; it only used Twitter data; and it lacked measures of redundancy in system output. This paper introduces Crisis Facts and Cross-Stream Temporal Summarization (CrisisFACTS), a new data challenge piloted in 2022 and developed to address these limitations. The CrisisFACTS framework recasts TREC-IS into an event-summarization task using multiple disaster-relevant data streams and a new fact-based evaluation scheme, allowing the community to assess state-of-the-art methods for summarizing disaster events Results from CrisisFACTS in 2022 include a new test-collection comprising human-generated disaster summaries along with multi-platform datasets of social media, crisis reports and news coverage for major crisis events.

Abstract: People increasingly use social media at the time of crisis, which produces a social media data deluge, where the public may find it difficult to locate trustworthy and credible information. Therefore, they often turn to authoritative sources: official individuals and organizations who are trusted to provide reliable information. It is then imperative that their credible messages reach and engage the widest possible audience, especially among those affected. In this study, we explore the role of metadata and linguistic factors in facilitating three types of engagement — retweets, replies, and favorites— with posts by authoritative sources. We find that many factors are similarly important across models (popularity, sociability, activity). However, some features are salient for only a specific type of engagement. We conclude by providing guidance to authoritative sources on how they may optimize specific types of engagement: retweets for information propagation, replies for in-depth sense-making, and favorites for cross-purpose visibility.

Abstract: Despite the outpouring of social support posted to social media channels in the aftermath of disaster, finding and managing content that can translate into community relief, donations, volunteering, or other recovery support is difficult due to the lack of sufficient annotated data around volunteerism. This paper outlines three experiments to alleviate these difficulties. First, we estimate to what degree volunteerism content from one crisis is transferable to another by evaluating the consistency of language in volunteer-and donation-related social media content across 78 disasters. Second it introduces methods for providing computational support in this emergency support function and developing semi-automated models for classifying volunteer-and donation-related social media content in new disaster events. Results show volunteer-and donation-related social media content is sufficiently similar across disasters and disaster types to warrant transferring models across disasters, and we evaluate simple resampling techniques for tuning these models. We then introduce and evaluate a weak-supervision approach to integrate domain knowledge from emergency response officers with machine learningmodelstoimproveclassification accuracy andacceleratethisemergencysupportinnewevents. This method helps to overcome the scarcity in data that we observe related to volunteer-and donation-related social media content.

Abstract: Over the years, with the increase in social media engagement, there has been an in increase in various pipelines to analyze, classify and prioritize crisis-related data on various social media platforms. These pipelines utilize various data augmentation methods to counter imbalanced crisis data, sophisticated and off-the-shelf models for training. However, there is a lack of comprehensive study which compares these methods for the various sections of a pipeline. In this study, we split a general crisis-related pipeline into 3 major sections, namely, data augmentation, model selection, and training methodology. We compare various methods for each of these sections and then present a comprehensive evaluation of which section to prioritize based on the results from various pipelines. We compare our results against two separate tasks, information classification and priority scoring for crisis-related tweets. Our results suggest that data augmentation, in general,improves the performance. However, sophisticated, state-of-the-art language models like DeBERTa only show performance gain in information classification tasks, and models like RoBERTa tend to show a consistent performance increase over our presented baseline consisting of BERT. We also show that, though training two separate task-specific BERT models does show better performance than one BERT model with multi-task learning methodology over an imbalanced dataset, multi-task learning does improve performance for more sophisticated model like DeBERTa with a much more balanced dataset after augmentation.

Abstract: Twitter is always worthwhile in facilitating communication during disasters. It helps in raising situational awareness and undertaking disaster control actions as quickly as possible to alleviate the miseries. But the noisy essence of Twitter causes difficulty in distinguishing relevant information from the heterogeneous contents. Therefore, extracting informative tweets is a substantial task to help in crisis intervention. Analyzing only the text or image content of the tweet often misses necessary insights which might be helpful during disasters. In this paper, we propose a multimodal framework to address the challenges of identifying informative crisis-related tweets containing both texts and images. Our presented approach incorporates an early fusion strategy of BERT-LSTM and ResNet50 networks which effectively learns from the joint representation of texts and images. The experiments and evaluation on the benchmark CrisisMMD dataset show that our fusion method surpasses the baseline by 7% and substantiates its potency over the unimodal systems.