Event-Related Bias Removal for Real-time Disaster Events

  title={Event-Related Bias Removal for Real-time Disaster Events},
  author={Evangelia Spiliopoulou and Salvador Medina Maza and Eduard H. Hovy and Alexander Hauptmann},
Social media has become an important tool to share information about crisis events such as natural disasters and mass attacks. Detecting actionable posts that contain useful information requires rapid analysis of huge volumes of data in real-time. This poses a complex problem due to the large amount of posts that do not contain any actionable information. Furthermore, the classification of information in real-time systems requires training on out-of-domain data, as we do not have any data from… 

Figures and Tables from this paper

Are Sudden Crises Making me Collapse? Measuring Transfer Learning Performances on Urgency Detection

The preliminary results are encouraging and show that transfer from predictable ecological crisis to sudden events is feasible and constitutes a first step towards real-time crisis management systems from social media content.

A Novel Framework for Detecting Important Subevents from Crisis Events via Dynamic Semantic Graphs

This work proposes a novel framework to extract critical sub-events from a large-scale crisis event by combining important information across relevant tweets and shows that this approach significantly outperforms event detection baselines.

Enhancing Crisis-Related Tweet Classification with Entity-Masked Language Modeling and Multi-Task Learning

This work found that entity-masking reduces the effect of overfitting to in-domain events and enables improvements in cross-event generalization in hierarchical multi-label classification as a multi-task learning problem.

Extracting a Knowledge Base of COVID-19 Events from Social Media

A manually annotated corpus of 10,000 tweets containing public reports of five COVID-19 events is presented, showing that the corpus can support fine-tuning BERT-based classifiers to automatically extract publicly reported events, which can be further collected for building a knowledge base.

UPB at SemEval-2021 Task 7: Adversarial Multi-Task Learning for Detecting and Rating Humor and Offense

This work describes their adversarial multi-task network, AMTL-Humor, used to detect and rate humor and offensive texts from Task 7 at SemEval-2021, and achieves a 95.66% F1-score and 94.70% accuracy.

Findings of the Association for Computational Linguistics: EMNLP 2020, Online Event, 16-20 November 2020

Unsupervised Few-Bits Semantic Hashing with Implicit Topics Modeling and Grid Tagging Scheme for End-to-End Fine-grained Opinion Extraction.

RESIN: A Dockerized Schema-Guided Cross-document Cross-lingual Cross-media Information Extraction and Event Tracking System

A new information extraction system that can automatically construct temporal event graphs from a collection of news documents from multiple sources, multiple languages, and multiple data modalities is presented.

Identifying informative messages in disaster events using Convolutional Neural Networks

This paper presents an approach to identifying informative messages in social media streams during disaster events based on Convolutional Neural Networks and shows significant improvement in performance over models that use the “bag of words” and n-grams as features on several datasets of messages from flooding events.

Applications of Online Deep Learning for Crisis Response Using Social Media Information

A new online algorithm based on stochastic gradient descent is proposed to train DNNs in an online fashion during disaster situations to address two types of information needs of response organizations: identifying informative tweets and classifying them into topical classes.

Few-shot tweet detection in emerging disaster events

This work compares how few-shot approaches (matching networks and prototypical networks) perform for this task and demonstrates how a modified one-class version of prototypical models can be used for this application.

Unconstrained Flood Event Detection Using Adversarial Data Augmentation

A new image classification framework based on adversarial data augmentation and deep learning algorithms to address the problem of flood event detection from images with real-world conditions is addressed.

Tweets Classification with BERT in the Field of Disaster Management

This project applies deep learning techniques to address Tweets classification problem in disaster management field, and shows that BERT and BERT-based LSTM attain the best results, outperforming the baseline model by 3.29% on average in terms of F-1 score respectively.

Twitter as a Lifeline: Human-annotated Twitter Corpora for NLP of Crisis-related Messages

This paper presents human-annotated Twitter corpora collected during 19 different crises that took place between 2013 and 2015, and publishes first largest word2vec word embeddings trained on 52 million crisis-related tweets.

TREC Incident Streams: Finding Actionable Information on Social Media

The current state-of-the-art emergency response technology is insufficient for emergency responders’ requirements, particularly for rare actionable information for which there is little prior training data available.

Crisis analytics: big data-driven crisis response

The history and the future of big crisis data analytics are introduced, along with a discussion on its promise, enabling technologies, challenges, and pitfalls.

Earthquake shakes Twitter users: real-time event detection by social sensors

This paper investigates the real-time interaction of events such as earthquakes in Twitter and proposes an algorithm to monitor tweets and to detect a target event and produces a probabilistic spatiotemporal model for the target event that can find the center and the trajectory of the event location.

Subword and Spatiotemporal Models for Identifying Actionable Information in Haitian Kreyol

A novel system, drawing on 40,000 emergency text messages sent in Haiti following the January 12, 2010 earthquake, predominantly in Haitian Kreyol, is presented, showing that keyword/ngram-based models using streaming MaxEnt achieve up to F=0.21 accuracy, and current state-of-the-art subword models increase this substantially.