Event Dataset Research Articles

Model tuning or prompt Tuning? a study of large language models for clinical concept and relation extraction

ObjectiveTo develop soft prompt-based learning architecture for large language models (LLMs), examine prompt-tuning using frozen/unfrozen LLMs, and assess their abilities in transfer learning and few-shot learning. MethodsWe developed a soft prompt-based learning architecture and compared 4 strategies including (1) fine-tuning without prompts; (2) hard-prompting with unfrozen LLMs; (3) soft-prompting with unfrozen LLMs; and (4) soft-prompting with frozen LLMs. We evaluated GatorTron, a clinical LLM with up to 8.9 billion parameters, and compared GatorTron with 4 existing transformer models for clinical concept and relation extraction on 2 benchmark datasets for adverse drug events and social determinants of health (SDoH). We evaluated the few-shot learning ability and generalizability for cross-institution applications. Results and ConclusionWhen LLMs are unfrozen, GatorTron-3.9B with soft prompting achieves the best strict F1-scores of 0.9118 and 0.8604 for concept extraction, outperforming the traditional fine-tuning and hard prompt-based models by 0.6 ∼ 3.1 % and 1.2 ∼ 2.9 %, respectively; GatorTron-345 M with soft prompting achieves the best F1-scores of 0.8332 and 0.7488 for end-to-end relation extraction, outperforming other two models by 0.2 ∼ 2 % and 0.6 ∼ 11.7 %, respectively. When LLMs are frozen, small LLMs have a big gap to be competitive with unfrozen models; scaling LLMs up to billions of parameters makes frozen LLMs competitive with unfrozen models. Soft prompting with a frozen GatorTron-8.9B model achieved the best performance for cross-institution evaluation. We demonstrate that (1) machines can learn soft prompts better than hard prompts composed by human, (2) frozen LLMs have good few-shot learning ability and generalizability for cross-institution applications, (3) frozen LLMs reduce computing cost to 2.5 ∼ 6 % of previous methods using unfrozen LLMs, and (4) frozen LLMs require large models (e.g., over several billions of parameters) for good performance.

Beyond Grounding: Extracting Fine-Grained Event Hierarchies across Modalities

Events describe happenings in our world that are of importance. Naturally, understanding events mentioned in multimedia content and how they are related forms an important way of comprehending our world. Existing literature can infer if events across textual and visual (video) domains are identical (via grounding) and thus, on the same semantic level. However, grounding fails to capture the intricate cross-event relations that exist due to the same events being referred to on many semantic levels. For example, the abstract event of "war'' manifests at a lower semantic level through subevents "tanks firing'' (in video) and airplane "shot'' (in text), leading to a hierarchical, multimodal relationship between the events. In this paper, we propose the task of extracting event hierarchies from multimodal (video and text) data to capture how the same event manifests itself in different modalities at different semantic levels. This reveals the structure of events and is critical to understanding them. To support research on this task, we introduce the Multimodal Hierarchical Events (MultiHiEve) dataset. Unlike prior video-language datasets, MultiHiEve is composed of news video-article pairs, which makes it rich in event hierarchies. We densely annotate a part of the dataset to construct the test benchmark. We show the limitations of state-of-the-art unimodal and multimodal baselines on this task. Further, we address these limitations via a new weakly supervised model, leveraging only unannotated video-article pairs from MultiHiEve. We perform a thorough evaluation of our proposed method which demonstrates improved performance on this task and highlight opportunities for future research. Data: https://github.com/hayyubi/multihieve

Characterizing Information Seeking Events in Health-Related Social Discourse

Social media sites have become a popular platform for individuals to seek and share health information. Despite the progress in natural language processing for social media mining, a gap remains in analyzing health-related texts on social discourse in the context of events. Event-driven analysis can offer insights into different facets of healthcare at an individual and collective level, including treatment options, misconceptions, knowledge gaps, etc. This paper presents a paradigm to characterize health-related information-seeking in social discourse through the lens of events. Events here are board categories defined with domain experts that capture the trajectory of the treatment/medication. To illustrate the value of this approach, we analyze Reddit posts regarding medications for Opioid Use Disorder (OUD), a critical global health concern. To the best of our knowledge, this is the first attempt to define event categories for characterizing information-seeking in OUD social discourse. Guided by domain experts, we develop TREAT-ISE, a novel multilabel treatment information-seeking event dataset to analyze online discourse on an event-based framework. This dataset contains Reddit posts on information-seeking events related to recovery from OUD, where each post is annotated based on the type of events. We also establish a strong performance benchmark (77.4% F1 score) for the task by employing several machine learning and deep learning classifiers. Finally, we thoroughly investigate the performance and errors of ChatGPT on this task, providing valuable insights into the LLM's capabilities and ongoing characterization efforts.

An event-oriented diffusion-refinement method for sparse events completion

Event cameras or dynamic vision sensors (DVS) record asynchronous response to brightness changes instead of conventional intensity frames, and feature ultra-high sensitivity at low bandwidth. The new mechanism demonstrates great advantages in challenging scenarios with fast motion and large dynamic range. However, the recorded events might be highly sparse due to either limited hardware bandwidth or extreme photon starvation in harsh environments. To unlock the full potential of event cameras, we propose an inventive event sequence completion approach conforming to the unique characteristics of event data in both the processing stage and the output form. Specifically, we treat event streams as 3D event clouds in the spatiotemporal domain, develop a diffusion-based generative model to generate dense clouds in a coarse-to-fine manner, and recover exact timestamps to maintain the temporal resolution of raw data successfully. To validate the effectiveness of our method comprehensively, we perform extensive experiments on three widely used public datasets with different spatial resolutions, and additionally collect a novel event dataset covering diverse scenarios with highly dynamic motions and under harsh illumination. Besides generating high-quality dense events, our method can benefit downstream applications such as object classification and intensity frame reconstruction.

An integrated approach for understanding global earthquake patterns and enhancing seismic risk assessment

Earthquakes, as intricate natural phenomena, profoundly impact lives, infrastructure, and the environment. While previous research has explored earthquake patterns through data analysis methods, there has been a gap in examining the time intervals between consecutive earthquakes across various magnitude categories. Given the complexity and vastness of seismic data, this study aims to provide comprehensive insights into global seismic activity by employing sophisticated data analysis methodologies on a century-long dataset of seismic events. The four-phase methodology encompasses exploratory data analysis (EDA), temporal dynamics exploration, spatial pattern analysis, and cluster analysis. The EDA serves as the foundational step, providing fundamental insights into the dataset's attributes and laying the groundwork for subsequent analyses. Temporal dynamics exploration focuses on discerning variations in earthquake occurrences over time. Spatial analysis identifies geographic regions with heightened earthquake activity and uncovers patterns of seismic clustering. K-means clustering is employed to delineate distinct earthquake occurrence clusters or hotspots based on geographical coordinates. The study's findings reveal a notable increase in recorded earthquakes since the 1960s, peaking in 2018. Distinct patterns in seismic activity are linked to factors such as time, human activities, and plate boundaries. The integrated approach enriches understanding of global earthquake trends and patterns, contributing to improved seismic hazard assessments, early warning systems, and risk mitigation efforts.

Biodiversity of zooplankton (Rotifera, Cladocera and Copepoda) in the tributaries of Cheboksary Reservoir (Middle Volga, Russia).

Freshwater zooplankton is an important component of the ecological communities of inland water bodies. It acts as an important part of the food web and participates in the self-purification processes of aquatic ecosystems. To study the abundance and distribution of species, a sampling event dataset was compiled and then published through GBIF. The aim of the work was to describe the current zooplankton fauna (Rotifera, Cladocera and Copepoda) and its abundance, based on a recently published dataset. The research was conducted from 2015 to 2022. Zooplankton samples were collected by vertical towing a plankton net (70 μm mesh) from the bottom to the water surface or by filtering through a net, the water being collected with a measuring bucket. The samples were concentrated to 100 ml and fixed with a final concentration of 4% formalin solution. For each sampling event, the coordinates of the location, number of individuals and date were recorded. The dataset contains information on 259 taxа, including 257 species and subspecies of zooplankton from 36 families found in the tributaries of the Cheboksary Reservoir. The families Chydoridae (35 species), Brachionidae (31) and Cyclopidae (27) were the most species-rich. Four invasive species were found: Kellicottiabostoniensis (Rousselet, 1908), Acanthocyclopsamericanus (Marsh, 1893), Ilyocryptusspinifer Herrick, 1882 and Thermocyclopstaihokuensis Harada, 1931.

War Dummies: Structured Data on Organised Armed Confrontations with Dutch Involvement, 1566–1812

Abstract The War dummies dataset offers structured data on Dutch-involved organised armed confrontations from 1566 to 1812. Comprising 1216 records detailing the participation of 95 entities in 548 encounters across 50 wars, it fills a crucial need for well-structured, accessible, and reusable pre-1815 historical warfare data. Based on the comprehensive Military History of the Netherlands book series, it aligns with post-1815 conflict datasets like the Inter-State War Database of the Correlates of War Project and the Georeferenced Event Dataset of the Uppsala Conflict Data Program. This article outlines the data collection, structure, and potential research applications, and discusses data quality and potential biases. The War dummies codebook offers comprehensive variable descriptions.

A Predictive Model of Vaccine Reactogenicity Using Data from an InVitro Human Innate Immunity Assay System.

A primary concern in vaccine development is safety, particularly avoiding an excessive immune reaction in an otherwise healthy individual. An accurate prediction of vaccine reactogenicity using invitro assays and computational models would facilitate screening and prioritization of novel candidates early in the vaccine development process. Using the modular invitro immune construct model of human innate immunity, PBMCs from 40 healthy donors were treated with 10 different vaccines of varying reactogenicity profiles and then cell culture supernatants were analyzed via flow cytometry and a multichemokine/cytokine assay. Differential response profiles of innate activity and cell viability were observed in the system. In parallel, an extensive adverse event (AE) dataset for the vaccines was assembled from clinical trial data. A novel reactogenicity scoring framework accounting for the frequency and severity of local and systemic AEs was applied to the clinical data, and a machine learning approach was employed to predict the incidence of clinical AEs from the invitro assay data. Biomarker analysis suggested that the relative levels of IL-1B, IL-6, IL-10, and CCL4 have higher predictive importance for AE risk. Predictive models were developed for local reactogenicity, systemic reactogenicity, and specific individual AEs. A forward-validation study was performed with a vaccine not used in model development, Trumenba (meningococcal group B vaccine). The clinically observed Trumenba local and systemic reactogenicity fell on the 26th and 93rd percentiles of the ranges predicted by the respective models. Models predicting specific AEs were less accurate. Our study presents a useful framework for the further development of vaccine reactogenicity predictive models.

Proposal for a new meteotsunami intensity index

Abstract. Atmospherically generated coastal waves labelled as meteotsunami are known to cause destruction, injury, and fatality due to their rapid onset and unexpected nature. Unlike other coastal hazards such as tsunami, there exist no standardised means of quantifying this phenomenon, which is crucial to understand shoreline impacts and to enable researchers to establish a shared language and framework for meteotsunami analysis and comparison. In this study, we present a new five-level Lewis Meteotsunami Intensity Index (LMTI) trialled in the United Kingdom (UK) but designed for global applicability. A comprehensive dataset of meteotsunami events recorded in the UK was utilised, and the index's effectiveness was evaluated, with intensity level and spatial distribution of meteotsunami occurrence derived. Results revealed a predominant occurrence of Level 2 moderate intensity meteotsunami (69 %) in the UK, with distinct hotspots identified in south-western England and Scotland. Further trial implementation of the LMTI in a global capacity revealed its potential adaptability to other meteotsunami-prone regions, facilitating the comparison of events and promoting standardisation of assessment methodologies.

An Events-Based Approach to Understanding Democratic Erosion

ABSTRACTThis article introduces and demonstrates the utility of a new event dataset on democratic erosion around the world. Through case studies of Turkey and Brazil, we show that our Democratic Erosion Event Dataset (DEED) can help to resolve debates about the extent to which democracy is backsliding based on prominent cross-national indicators, focusing in particular on the Varieties of Democracy (V-Dem) and Little and Meng (L&M) indices. V-Dem suggests that democracies are deteriorating worldwide; L&M argue that this may be an artifact of subjectivity and coder bias and that more “objective” indicators reveal little to no global democratic backsliding in recent years. Using DEED, we show that—at least in these cases—objective indices may underestimate the extent of democratic erosion whereas subjective indices may overestimate it. Our analyses illustrate the ways in which DEED can complement existing indices by illuminating the nature and dynamics of democratic erosion as it occurs on the ground.

Meta-Analysis and the Integration of Terrorism Event Databases

Why do terrorist attacks occur in certain places and times but not others? Despite advances in collection and empirical methods, the literature has produced divergent results and reached little consensus for common hypotheses about the economic, political, and social causes of terrorism. It is hard to know what to make disagreements as studies adopt disparate research designs using different datasets covering different locations and times. This article applies the xSub data protocol to conduct a meta-analysis of terrorism event datasets and isolate explanations for variations in findings. Although the datasets are constructed for different purposes by different research teams, with different inclusion standards, processing data onto a common event typology, and conducting analysis across common coverage reduces heterogeneity in findings. This protocol also facilitates comparisons with general conflict event datasets, providing researchers, policymakers, and practitioners with a broader context for understanding terrorism in relation to other forms of violence.

An Emotional Analysis of Korean Topics based on Social Media Big Data Clustering

An innovative approach is introduced in this paper to address the challenges in emotional topic interpretation and accuracy in emotional situation assessment. Utilizing large data from social media to improve the accuracy of emotional analysis in online debates, with a specific emphasis on Korean themes. The proposed solution, the Online Topic Emotion Recognition Model (OTSRM), builds upon the foundational Online Latent Dirichlet Allocation (OLDA) model. The OTSRM integrates the concept of emotion intensity and introduces an inventive emotion iteration framework to tackle these issues. Key innovations of the OTSRM include establishing an affective evolution channel by augmenting affective heritability using a β priori. Additionally, the model generates two critical distribution matrices: one for characteristic words and another for affective words, facilitating a deeper understanding of emotional context within topics. The relative entropy method is employed to discern emotional tones in textual content, calculating maximum emotion values for topic focus within adjacent time segments. Validation experiments using five diverse network event datasets and comparisons to mainstream models demonstrate the OTSRM's effectiveness with emotion recognition accuracy rates of 85.56% and 81.03%. The OTSRM represents significant progress in addressing challenges associated with emotional topic analysis and precise emotional dynamics assessment in Korean social media data.

Context-Aware Proposal-Boundary Network With Structural Consistency for Audiovisual Event Localization.

Audiovisual event localization aims to localize the event that is both visible and audible in a video. Previous works focus on segment-level audio and visual feature sequence encoding and neglect the event proposals and boundaries, which are crucial for this task. The event proposal features provide event internal consistency between several consecutive segments constructing one proposal, while the event boundary features offer event boundary consistency to make segments located at boundaries be aware of the event occurrence. In this article, we explore the proposal-level feature encoding and propose a novel context-aware proposal-boundary (CAPB) network to address audiovisual event localization. In particular, we design a local-global context encoder (LGCE) to aggregate local-global temporal context information for visual sequence, audio sequence, event proposals, and event boundaries, respectively. The local context from temporally adjacent segments or proposals contributes to event discrimination, while the global context from the entire video provides semantic guidance of temporal relationship. Furthermore, we enhance the structural consistency between segments by exploiting the above-encoded proposal and boundary representations. CAPB leverages the context information and structural consistency to obtain context-aware event-consistent cross-modal representation for accurate event localization. Extensive experiments conducted on the audiovisual event (AVE) dataset show that our approach outperforms the state-of-the-art methods by clear margins in both supervised event localization and cross-modality localization.

ERDAP: A Novel Method of Event Relation Data Augmentation Based on Relation Prediction

Event relation extraction is a key aspect in the fields of event evolutionary graph construction, knowledge question and answer, and intelligence analysis, etc. Currently, supervised learning methods that rely on large amounts of labeled data are mostly used; however, the size of existing event relation datasets is small and cannot provide sufficient training data for the models. To alleviate this challenging research question, this study proposes a novel data augmentation model, called Event Relation Data Augmentation based on relationship Prediction (ERDAP), that allows both semantic and structural features to be taken into account without changing the semantic relation label compatibility, uses event relation graph convolutional neural networks to predict event relations, and expands the generated high-quality event relation triples as new training data for the event relation texts. Experimental evaluation using event causality extraction method on Chinese emergent event dataset shows that our model significantly outperforms existing text augmentation methods and achieves desirable performance, which provides a new idea for event relation data augmentation

Bayesian High-Rank Hankel Matrix Completion for Nonlinear Synchrophasor Data Recovery

Phasor measurement units (PMUs) provide high temporal-resolution synchrophasor measurements for power system monitoring and control. The frequent data quality issues, such as missing and bad data, prevent the incorporation of synchrophasor data in real-time operations. Most existing data-driven data recovery methods assume the power system dynamics can be approximated by a linear dynamical system, and the recovery performance degrades significantly when the power system is experiencing nonlinear dynamics during significant events. This paper proposes a data-driven Bayesian nonlinear synchrophasor data recovery method (Ba-NSDR) that can recover a consecutive time period of simultaneous data losses or errors across all channels, even when the underlying system is highly nonlinear. The idea is to lift the Hankel matrix of the spatial-temporal synchrophasor data to a higher dimension such that the lifted Hankel matrix is low-rank in that space and can be processed with the kernel trick. Our proposed Bayesian method then infers the probabilistic distributions of synchrophasor from the partial observations. Some distinctive features of Ba-NSDR include an uncertainty index to measure the accuracy of the recovery result and the robustness to parameter selections. Our method is verified on both synthetic and recorded event datasets.

Backlight and dim space object detection based on a novel event camera.

For space object detection tasks, conventional optical cameras face various application challenges, including backlight issues and dim light conditions. As a novel optical camera, the event camera has the advantages of high temporal resolution and high dynamic range due to asynchronous output characteristics, which provides a new solution to the above challenges. However, the asynchronous output characteristic of event cameras makes them incompatible with conventional object detection methods designed for frame images. Asynchronous convolutional memory network (ACMNet) for processing event camera data is proposed to solve the problem of backlight and dim space object detection. The key idea of ACMNet is to first characterize the asynchronous event streams with the Event Spike Tensor (EST) voxel grid through the exponential kernel function, then extract spatial features using a feed-forward feature extraction network, and aggregate temporal features using a proposed convolutional spatiotemporal memory module ConvLSTM, and finally, the end-to-end object detection using continuous event streams is realized. Comparison experiments among ACMNet and classical object detection methods are carried out on Event_DVS_space7, which is a large-scale space synthetic event dataset based on event cameras. The results show that the performance of ACMNet is superior to the others, and the mAP is improved by 12.7% while maintaining the processing speed. Moreover, event cameras still have a good performance in backlight and dim light conditions where conventional optical cameras fail. This research offers a novel possibility for detection under intricate lighting and motion conditions, emphasizing the superior benefits of event cameras in the realm of space object detection.

“THIS TWEET IS UNAVAILABLE”: #BLACKLIVESMATTER TWEETS DECAY

Previous studies show that random tweet collections that include more than one hashtag had relatively low rate of unavailable tweets while political related datasets have a higher rate of tweet decay (Almuhimedi et al., 2013; Bastos, 2021; Bhattacharya & Ganguly, 2016; McCammon, 2022; Zubiaga, 2018). For example, Zubiaga’s (2018) study indicates that 81.4% of their 30 randomly selected real-world events datasets remained available after 4 years, while Bastos’ (2021) study shows that only 67% of Brexit debate related tweets were available after the same amount of time. Our preliminary study looks at the #BlackLivesMatter discussion on Twitter and we find that only 63% tweets remain available after 2 years, which is significantly higher compared to the previous studies. This preliminary work adds to the existing literature in two keyways. First, we broaden the kinds of discussion spaces where information decay has been studied by focusing on #BlackLivesMatter (hereafter, BLM), which in the U.S. has become a highly politicized movement (Stewart et al., 2018). Second, to our best knowledge, this is the first study looking at social movement tweets since Musk took over and it may highlight how the landscape on Twitter has shifted.

A novel attention model across heterogeneous features for stuttering event detection

Stuttering is a prevalent speech disorder affecting millions worldwide. To provide an automatic and objective stuttering assessment tool, Stuttering Event Detection (SED) is under extensive investigation for advanced speech research and applications. Despite significant progress achieved by various machine learning and deep learning models, SED directly from speech signal is still challenging due to stuttering speech’s heterogeneous and overlapped nature. This paper presents a novel SED approach using multi-feature fusion and attention mechanisms. The model utilises multiple acoustic features extracted based on different pitch, time-domain, frequency domain, and automatic speech recognition feature to detect stuttering core behaviours more accurately and reliably. In addition, we exploit both spatial and temporal attention mechanisms as well as Bidirectional Long Short-Term Memory (BI-LSTM) modules to learn better representations to improve the SED performance. The experimental evaluation and analysis convincingly demonstrate that our proposed model surpasses the state-of-the-art models on two popular stuttering datasets, with 4% and 3% overall F1 scores, respectively. The superior results indicate the consistency of our proposed method, supported by both multi-feature and attention mechanisms in different stuttering events datasets.

‘Serving other oppressed peoples’: the Radical Translations project and the extension of liberty (1789–1815)

The Radical Translations Project considers the role of activist translators in extending radical, democratic ideas into new contexts in the period 1789–1815. It uses both bibliographical and prosopographical data-models to construct a corpus of c.1000 ‘activist’ translations and linked biographies of some 500 translators working across English, French and Italian languages. This article discusses the challenges of building a multilingual digital corpus in the absence of any pre-existing digital or analogue collection. It also considers the relevance of network analysis to prosopography at a time when many translators wrote anonymously or pseudonymously. Finally it considers how digital tools can be used to cross-reference small-scale but highly granular datasets of texts, people, places and events. Along the way, we comment on the role of selection in the construction of any database and the importance of making this (irreducibly subjective) process more visible to users.

Plots, Attacks, and the Measurement of Terrorism

How should we measure terrorism? Political scientists typically use executed attacks as the dependent variable and test covariates to identify factors that produce terrorism. But attacks are an imperfect measure of terrorist activity because of ‘plot attrition’ — the tendency for plots to derail due to police intervention or other factors. We examine whether the exclusion of foiled plots from event datasets constitutes a measurement problem in terrorism studies. Building on recent advances in plot data collection, we study the correlation between plots and attacks and conduct an original analysis of jihadism in Europe. Our results suggest common research designs predicting terrorism can produce different results depending on whether incidents are operationalized as plots or attacks. Adjusting for state security capability does not solve the problem. Despite its limitations, plot data is a more complete measure of terrorist activity that should be incorporated, when available, in quantitative studies of terrorism.