Abstract: During the course of this pandemic, the use of social media and virtual networks has been at an all-time high. Individuals have used social media to express their thoughts on matters related to this pandemic. It is difficult to predict current trends based on historic case data because trends are more connected to social activities which can lead to the spread of coronavirus. So, it's important for us to derive meaningful information from social media as it is widely used. Therefore, we grouped tweets by common keywords, found correlations between keywords and daily COVID-19 statistics and built predictive modeling. The features correlation analysis was very effective, so trends were predicted very well. A RMSE score of 0.0425504, MAE of 0.03295105 and RSQ of 0.5237014 in relation to daily deaths. In addition, we found a RMSE score of 0.07346836, MAE of 0.0491152 and RSQ 0.374529 in relation to daily cases.

Abstract: Between 2018 and 2019, the Incident Streams track (TREC-IS) has developed standard approaches for classifying the types and criticality of information shared in online social spaces during crises, but the introduction of SARS-CoV-2 has shifted the landscape of online crises substantially. While prior editions of TREC-IS have lacked data on large-scale public-health emergencies as these events are exceedingly rare, COVID-19 has introduced an over-abundance of potential data, and significant open questions remain about how existing approaches to crisis informatics and datasets built on other emergencies adapt to this new context. This paper describes how the 2020 edition of TREC-IS has addressed these dual issues by introducing a new COVID-19-specific task for evaluating generalization of existing COVID-19 annotation and system performance to this new context, applied to 11 regions across the globe. TREC-IS has also continued expanding its set of target crises, adding 29 new events and expanding the collection of event types to include explosions, fires, and general storms, making for a total of 9 event types in addition to the new COVID-19 events. Across these events, TREC-IS has made available 478,110 COVID-related messages and 282,444 crisis-related messages for participant systems to analyze, of which 14,835 COVID-related and 19,784 crisis-related messages have been manually annotated. Analyses of these new datasets and participant systems demonstrate first that both the distributions of information type and priority of information vary between general crises and COVID-19-related discussion. Secondly, despite these differences, results suggest leveraging general crisis data in the COVID-19 context improves performance over baselines. Using these results, we provide guidance on which information types appear most consistent between general crises and COVID-19.

Abstract: Social media has enabled people to circulate information in a timely fashion, thus motivating people to post messages seeking help during crisis situations. These messages can contribute to the situational awareness of emergency responders, who have a need for them to be categorised according to information types (i.e. the type of aid services the messages are requesting). We introduce a transformer-based multi-task learning (MTL) technique for classifying information types and estimating the priority of these messages. We evaluate the effectiveness of our approach with a variety of metrics by submitting runs to the TREC Incident Streams (IS) track: a research initiative specifically designed for disaster tweet classification and prioritisation. The results demonstrate that our approach achieves competitive performance in most metrics as compared to other participating runs. Subsequently, we find that an ensemble approach combining disparate transformer encoders within our approach helps to improve the overall effectiveness to a significant extent, achieving state-of-the-art performance in almost every metric. We make the code publicly available so that our work can be reproduced and used as a baseline for the community for future work in this domain.

Abstract: Twitter is an immediate and almost ubiquitous platform and therefore can be a valuable source of information during disasters. Current methods for identifying and classifying crisis-related content are often based on single tweets, i.e., already known information from the past is neglected. In this paper, the combination of tweet-wise pre-trained neural networks and unsupervised semantic clustering is proposed and investigated. The intention is to (1) enhance the generalization capability of pre-trained models, (2) to be able to handle massive amounts of stream data, (3) to reduce information overload by identifying potentially crisis-related content, and (4) to obtain a semantically aggregated data representation that allows for further automated, manual and visual analyses. Latent representations of each tweet based on pre-trained sentence embedding models are used for both, clustering and tweet classification. For a fast, robust and time-continuous processing, subsequent time periods are clustered individually according to a Chinese restaurant process. Clusters without any tweet classified as crisis-related are pruned. Data aggregation over time is ensured by merging semantically similar clusters. A comparison of our hybrid method to a similar clustering approach, as well as first quantitative and qualitative results from experiments with two different labeled data sets demonstrate the great potential for crisis-related Twitter stream analyses.

Abstract: Social media applications such as Twitter and Facebook are fast becoming a key instrument in gaining situational awareness (understanding the bigger picture of the situation) during disasters. This has provided multiple opportunities to gather relevant information in a timely manner to improve disaster response. In recent years, identifying crisis-related social media posts is analysed as an automatic task using machine learning (ML) or deep learning (DL) techniques. However, such supervised learning algorithms require labelled training data in the early hours of a crisis. Recently, multiple manually labelled disaster-related open-source twitter datasets have been released. In this work, we create a large dataset with 186,718 tweets by combining a number of such datasets and evaluate the performance of multiple ML and DL algorithms in classifying disaster-related tweets in three settings, namely ``in-disaster'', ``out-disaster'' and ``cross-disaster''. Our results show that the Bidirectional LSTM model with Word2Vec embeddings performs well for the tweet classification task in all three settings. We also make available the preprocessing steps and trained weights for future research.

Abstract: In 2020, we have witnessed an unprecedented crisis event, the COVID-19 pandemic. Various questions arise regarding the nature of this crisis data and the impacts it would have on the existing tools. In this paper, we aim to study whether we can include pandemic-type crisis events with general non-pandemic events and hypothesize that including labeled crisis data from a variety of non-pandemic events will improve classification performance over models trained solely on pandemic events. To test our hypothesis we study the model performance for different models by performing a cross validation test on pandemic only held-out sets for two different types of training sets, one containing only pandemic data and the other a combination of pandemic and non-pandemic crisis data, and comparing the results of the two. Our results approve our hypothesis and give evidence of some crucial information propagation upon inclusion of non-pandemic crisis data to pandemic data.

Abstract: In this paper, we show that the text-image relationship of disaster tweets can be used to improve the classification of tweets from emergency situations. To this end, we introduce DisRel, a dataset which contains 4,600 multimodal tweets, collected during the disasters that hit the USA in 2017, and manually annotated with coherence image-text relationships, such as Similar and Complementary. We explore multiple models to detect these relationships and perform a comprehensive analysis into the robustness of these methods. Based on these models, we build a simple feature augmentation approach that can leverage the text-image relationship. We test our methods on 2 tasks in CrisisMMD: Humanitarian Categories and Damage Assessment, and observe an increase in the performance of the relationship-aware methods.

Abstract: This paper summarizes key opportunities and challenges identified during the workshop “Social Media for Disaster Risk Management: Researchers Meet Practitioners” which took place online in November 2020. It constitutes a work-in-progress towards identifying new directions for research and development of systems that can better serve the information needs of emergency managers. Practitioners widely recognize the potential of accessing timely information from social media. Nevertheless, the discussion outlined some critical challenges for improving its adoption during crises. In particular, validating such information and integrating it with authoritative information and into more traditional information systems for emergency managers requires further work, and the negative impacts of misinformation and disinformation need to be prevented.

Abstract: Timely and reliable sensing of infrastructure conditions is critical in disaster management for planning effective infrastructure restorations. Social media, a near real-time information source, has been widely used in the disaster domain for building timely, general situational awareness, such as urgent public needs and donations. However, the employment of social media for sensing electricity infrastructure conditions has yet been explored. This study aims to address the research gap to sense electricity infrastructure conditions through mining public topics from social media. To achieve this purpose, we proposed a systematic and customized approach wherein (1) electricity-related social media data is extracted by the classifier developed based on Bidirectional Encoder Representations from Transformers (BERT); and (2) public topics are modeled with unigrams, bigrams, and trigrams to incorporate the formulaic expressions of infrastructure conditions in social media. Electricity infrastructures in Florida impacted by Hurricane Irma are studied for illustration and demonstration. Results show that the proposed approach is capable of sensing the temporal evolutions and geographic differences of electricity infrastructure conditions.