Papers by Cornelia Caragea

Semi-Supervised Domain Adaptation for Emotion-Related Tasks (2023.findings-acl)

Challenge:	Recent studies show improvements in stance detection by using attention mechanism or sentiment information.
Approach:	They propose a multi-task framework that incorporates attention mechanism and takes sentiment classification as an auxiliary task.
Outcome:	The proposed model outperforms state-of-the-art deep learning methods on the SemEval-2016 dataset.

Copied to clipboard

Mahshid Hosseini, Cornelia Caragea

Low-Resource NLP

Challenge:	Semi-supervised domain adaptation (SSDA) is a model trained from a label-rich source domain to a new but related domain with a few labels of target data.
Approach:	They propose to decompose the semi-supervised domain adaptation framework into two subcomponents of unsupervised domain adaption (UDA) from the source to the target domain and semi-supervised learning (SSL) in the target.
Outcome:	The proposed method is based on the co-learning of multiple classifiers for computer vision tasks and is published in the journal Nature.

On the Calibration of Pre-trained Language Models using Mixup Guided by Area Under the Margin and Saliency (2022.acl-long)

Copied to clipboard

Seo Yeon Park, Cornelia Caragea

EnsyNet: A Dataset for Encouragement and Sympathy Detection (2022.lrec-1)

Challenge:	Existing studies have shown that mixing up can improve model calibration on image classification tasks, but little is known about using it on natural language understanding (NLU) tasks.
Approach:	They propose a mixup strategy for pre-trained language models that improves model calibration further by using the AUM statistic and saliency map.
Outcome:	The proposed mixup improves model calibration on natural language understanding tasks while maintaining competitive accuracy.

Copied to clipboard

Stance Detection in COVID-19 Tweets (2021.acl-long)

Challenge:	Existing studies on emotional support detection focus on the presence or absence of emotional support, while the available datasets are limited or scarce in terms of size.
Approach:	They propose to use a dataset of 6,500 sentences annotated with encouragement and sympathy to train BERT-based classifiers on this dataset and apply their best BERT model to two large scale experiments.
Outcome:	The proposed model improves the emotional state of users while the lack of emotional support negatively impacts patients’ emotional state.

Copied to clipboard

Kyle Glandt, Sarthak Khanal, Yingjie Li, Doina Caragea, Cornelia Caragea

Cross-Lingual Disaster-related Multi-label Tweet Classification with Manifold Mixup (2020.acl-srw)

Challenge:	a global pandemic of COVID-19 has forced major changes in our daily lives . a new stance detection dataset is being used to track the stances of Twitter users .
Approach:	They use Twitter stance data to collect stances on topics related to the pandemic . they train models to take advantage of large amounts of unlabeled data .
Outcome:	The proposed model improves on existing stance detection datasets and unlabeled data.

Copied to clipboard

Jishnu Ray Chowdhury, Cornelia Caragea, Doina Caragea

JointMatch: A Unified Approach for Diverse and Collaborative Pseudo-Labeling to Semi-Supervised Text Classification (2023.emnlp-main)

Challenge:	Towards this goal, many studies have focused on disaster-related tweet classification.
Approach:	They compile a multilingual dataset for multi-label classification of disaster-related tweets . they show that their model generalizes to unseen disasters in the test set .
Outcome:	The proposed model generalizes to unseen disasters and improves with Manifold Mixup.

Copied to clipboard

Henry Zou, Cornelia Caragea

eMLM: A New Pre-training Objective for Emotion Related Tasks (2021.acl-short)

Challenge:	Existing approaches to semi-supervised text classification suffer from pseudo-label bias and error accumulation.
Approach:	They propose a pseudo-labeling approach to semi-supervised text classification that unifies ideas from semi-semi-supervised learning and the task of learning with noise.
Outcome:	The proposed approach achieves a significant improvement on benchmark datasets even in the extremely-scarce-label setting.

Copied to clipboard

Exploring Optimism and Pessimism in Twitter Using Deep Learning (D18-1)

Challenge:	Emotion Masked Language Modelling improves the performance of a pretraining language model for emotion detection and sentiment analysis tasks.
Approach:	They propose a BERT-based version of Masked Language Modelling that induces emotion into the model.
Outcome:	The proposed model improves on emotion detection and sentiment analysis tasks by 1.2% F-1 . the proposed model also shows increased robustness in the test.

Copied to clipboard

Cornelia Caragea, Liviu P. Dinu, Bogdan Dumitru

MultiMatch: Multihead Consistency Regularization Matching for Semi-Supervised Text Classification (2025.emnlp-main)

Challenge:	Identifying optimistic and pessimistic viewpoints and users from Twitter is useful for providing better social support to those who need it.
Approach:	They propose deep learning models to predict optimism and pessimism in Twitter . they also show that a sentiment classifier would not be sufficient for predicting optimism and psi .
Outcome:	The proposed models outperform traditional machine learning classifiers on optimism and pessimism in Twitter.

Copied to clipboard

Iustin Sirbu, Robert-Adrian Popovici, Cornelia Caragea, Stefan Trausan-Matu, Traian Rebedea

Dynamic Classification in Web Archiving Collections (2020.lrec-1)

Challenge:	MultiMatch is a semi-supervised learning (SSL) algorithm that combines co-training and consistency regularization with pseudo-labeling.
Approach:	They propose a semi-supervised learning algorithm that integrates co-training and consistency regularization with pseudo-labeling.
Outcome:	The proposed algorithm outperforms the second-best approach on 8 out of 10 setups from 5 natural language processing datasets and outperformed the second best by 3.26%.

Copied to clipboard

Krutarth Patel, Cornelia Caragea, Mark Phillips

Hard Emotion Test Evaluation Sets for Language Models (2025.findings-naacl)

Challenge:	a growing number of research libraries, museums, and archives are embracing Web Archiving as a mechanism to collect born-digital material made available via the Web.
Approach:	They propose to use dynamic fusion models to find the model that performs best on a variety of document types.
Outcome:	The proposed model outperforms individual models and other ensemble methods on three datasets.

Copied to clipboard

Multimodal Semi-supervised Learning for Disaster Tweet Classification (2022.coling-1)

Challenge:	Existing tests on emotion datasets do not show whether language models understand emotions or exploit supperficial lexical cues.
Approach:	They propose to use two existing emotion datasets to evaluate whether language models make inferential decisions for emotion detection.
Outcome:	The proposed test sets evaluate language models on emotion datasets.

Copied to clipboard

Iustin Sirbu, Tiberiu Sosea, Cornelia Caragea, Doina Caragea, Traian Rebedea

Scaling Performance and Low-Resource Annotation with Many-Shot In-Context Learning for Named Entity Recognition (2026.findings-acl)

Challenge:	During natural disasters, people use social media platforms to post information about casualties and damage . annotating data can be burdensome, subjective and expensive . et al., 2018b; sohn e.t., 2020) proposed semi-supervised multimodal approach to improve performance on multimodal tasks.
Approach:	They propose a semi-supervised approach to annotate unlabeled data from Twitter . they extend FixMatch algorithm to a multimodal setting to account for subjective data .
Outcome:	The proposed approach improves on multimodal disaster tweet classification tasks.

Copied to clipboard

Qi Zhang, Fangping Lan, Cornelia Caragea, Longin Jan Latecki, Eduard Dragut

Low-Resource NLP

Challenge:	Existing studies on ICL for Named Entity Recognition (NER) have mainly explored few-shot settings, but the potential of scaling to hundreds of demonstrations has not been thoroughly investigated.
Approach:	They evaluate various LLMs across multiple domains using hundreds of ICL examples and then assess the feasibility of using many-shot ICL as a data annotation framework.
Outcome:	The proposed framework can be scaled to hundreds of examples and annotate and refining data for low-resource NER tasks.

SciER: An Entity and Relation Extraction Dataset for Datasets, Methods, and Tasks in Scientific Documents (2024.emnlp-main)

Copied to clipboard

Qi Zhang, Zhijia Chen, Huitong Pan, Cornelia Caragea, Longin Latecki, Eduard Dragut

Towards Identifying Fine-Grained Depression Symptoms from Memes (2023.acl-long)

Challenge:	Scientific information extraction (SciIE) is critical for converting unstructured knowledge from scholarly articles into structured data.
Approach:	They propose to use a scientific entity and relation extraction dataset to capture interactions between entities in full texts.
Outcome:	The proposed dataset captures the intricate use and interactions among entities in full texts and provides an out-of-distribution test set to offer a more realistic evaluation.

Copied to clipboard

Shweta Yadav, Cornelia Caragea, Chenye Zhao, Naincy Kumari, Marvin Solberg, Tanmay Sharma

Multimodality

Challenge:	Mental health disorders are a major economic burden for society and are projected to rise to a staggering US $6 trillion by 2030.
Approach:	They propose to use memes to identify fine-grained depression symptoms from memes . they benchmark RESTORE on 20 strong monomodal and multimodal methods .
Outcome:	The proposed method can predict fine-grained depression symptoms better than existing models that overlook implicit connections between visual and textual elements of a meme.

Improving Stance Detection with Multi-Dataset Learning and Knowledge Distillation (2021.emnlp-main)

Copied to clipboard

Yingjie Li, Chenye Zhao, Cornelia Caragea

KPDROP: Improving Absent Keyphrase Generation (2022.findings-emnlp)

Challenge:	stance detection is a method to determine whether a text author is in favor of, against or neutral toward a specific target.
Approach:	They propose a method that applies instance-specific temperature scaling to the teacher and student predictions.
Outcome:	The proposed method outperforms the state-of-the-art on all datasets and on multiple datasets.

Copied to clipboard

Jishnu Ray Chowdhury, Seo Yeon Park, Tuhin Kundu, Cornelia Caragea

Text Generation

Challenge:	Existing approaches to generate absent keyphrases have limited results in generating present keyphrase.
Approach:	They propose a model-agnostic approach to generate absent keyphrases by randomly dropping present keyphrase from a document and turning them into artificial absent key phrases during training.
Outcome:	The proposed approach improves absent keyphrase generation in both supervised and resource-constrained settings.

Distilling Calibrated Knowledge for Stance Detection (2023.findings-acl)

Copied to clipboard

A Data Cartography based MixUp for Pre-trained Language Models (2022.naacl-main)

Challenge:	Existing methods for stance detection ignore meaningful signals among categories offered by hard labels.
Approach:	They propose to use knowledge distillation to calibrate teacher predictions in each generation step.
Outcome:	The proposed method can calibrate teacher predictions in each generation step and improves stance detection accuracy.

Copied to clipboard

Seo Yeon Park, Cornelia Caragea

Hierarchical Multi-Label Classification of Scientific Documents (2022.emnlp-main)

Challenge:	Using MixUp, additional samples are generated during training by combining random pairs of training samples and their labels.
Approach:	They propose a new MixUp strategy that leverages Training Dynamics and allows more informative samples to be combined for generating new data samples.
Outcome:	The proposed method achieves competitive performance using a smaller subset of training data compared with strong baselines and yields lower expected calibration error on the pre-trained language model, BERT, on both in-domain and out-of-domain settings.

Copied to clipboard

How Hard is this Test Set? NLI Characterization by Exploiting Training Dynamics (2024.emnlp-main)

Challenge:	Automated topic classification is a useful tool for managing scientific documents in a digital collection.
Approach:	They propose a hierarchical multi-label text classification dataset with keyword labeling as an auxiliary task.
Outcome:	The proposed model achieves a Macro-F1 score of 34.57% and is publicly available.

Copied to clipboard

Adrian Cosma, Stefan Ruseti, Mihai Dascalu, Cornelia Caragea

Challenge:	Popular datasets suffer from systematic spurious correlations that artificially inflate actual model performance.
Approach:	They propose a method for the automated creation of a challenging test set without relying on manual construction of artificial and unrealistic examples.
Outcome:	The proposed method reduces spurious correlations and improves model performance . examples labeled as having the highest difficulty show markedly decreased performance compared to the full dataset .

Unsupervised Extractive Summarization of Emotion Triggers (2023.acl-long)

Copied to clipboard

Tiberiu Sosea, Hongli Zhan, Junyi Jessy Li, Cornelia Caragea

A New Public Corpus for Clinical Section Identification: MedSecId (2022.coling-1)

Challenge:	Recent approaches trained supervised models to detect emotions and explain emotion triggers via abstractive summarization, but this can block necessary responses.
Approach:	They propose to augment an abstractive dataset with extractive triggers and develop unsupervised models that can jointly detect emotions and summarize their triggers.
Outcome:	The proposed model outperforms existing models and is based on a COVID-19 crisis dataset.

Copied to clipboard

Paul Landes, Kunal Patel, Sean S. Huang, Adam Webb, Barbara Di Eugenio, Cornelia Caragea

Challenge:	a study aims to segment sections of clinical medical domain documentation . section identification is a process by which sections are demarcated and labeled .
Approach:	They use a set of 2,002 fully annotated medical notes from the MIMIC-III to segment sections in clinical medical domain documentation.
Outcome:	The proposed model shows that medical concepts are related across sections using principal component analysis.

DynClean: Training Dynamics-based Label Cleaning for Distantly-Supervised Named Entity Recognition (2025.findings-naacl)

Copied to clipboard

Qi Zhang, Huitong Pan, Zhijia Chen, Longin Jan Latecki, Cornelia Caragea, Eduard Dragut

ZeroStance: Leveraging ChatGPT for Open-Domain Stance Detection via Dataset Generation (2024.findings-acl)

Challenge:	Existing methods to identify entities using distant annotations are expensive and time-consuming.
Approach:	They propose a training dynamics-based label cleaning approach to characterize distant annotations and an automatic threshold estimation strategy to locate errors in distant labels.
Outcome:	The proposed method outperforms several advanced DS-NER approaches across four datasets.

Copied to clipboard

Chenye Zhao, Yingjie Li, Cornelia Caragea, Yue Zhang

Target-Aware Data Augmentation for Stance Detection (2021.naacl-main)

Challenge:	Until recently, zero-shot stance detection was limited to in-domain tasks.
Approach:	They propose a method for stance detection that trains a model that can generalize well to unseen targets across multiple domains.
Outcome:	The proposed method generalizes well to unseen targets across multiple domains over baselines on most benchmarks.

Copied to clipboard

Leveraging Training Dynamics and Self-Training for Text Classification (2022.findings-emnlp)

Challenge:	Existing methods for stance detection are not diversified or inconsistent with the given target and label information.
Approach:	They propose to augment a text with a conditional masked word prediction task . they propose to replace a target mention with 'target-aware' sentences by replacing a reference word with .
Outcome:	The proposed method outperforms existing methods on 11 targets.

Copied to clipboard

A Multi-Task Learning Framework for Multi-Target Stance Detection (2021.findings-acl)

Text Classification

Challenge:	Semi-supervised learning (SSL) is a promising technique for improving deep learning models when training data is scarce.
Approach:	They propose a semi-supervised learning approach that leverages training dynamics of unlabeled data.
Outcome:	The proposed method achieves an average increase in F1 score of 3.5% over baselines in low resource settings.

Copied to clipboard

Zero-Shot Keyphrase Generation: Investigating Specialized Instructions and Multi-sample Aggregation on Large Language Models (2025.findings-naacl)

Low-Resource NLP

Challenge:	Existing models fail to learn target-specific representations and are prone to overfitting.
Approach:	They propose a multi-task learning network to train one model on all target pairs . their results show that their proposed model outperforms the best-performing baseline by 12.39% .
Outcome:	The proposed model outperforms the best-performing baseline model by 12.39% in macro-averaged F1-score.

Copied to clipboard

Jishnu Ray Chowdhury, Jayanth Mohan, Tomas Malik, Cornelia Caragea

MADIAVE: Multi-Agent Debate for Implicit Attribute Value Extraction (2026.findings-eacl)

Challenge:	Keyphrase generation is a long-standing NLP task for automatically generating keyphrases for a given document.
Approach:	They propose to use open-source instruction-tuned LLMs for keyphrase generation . they propose task-specific counterparts to self-consistency-style strategies for LLM .
Outcome:	The proposed model improves on existing models and shows significant benefits over baselines.

Copied to clipboard

Wei-Chieh Huang, Cornelia Caragea

Multimodality

Challenge:	Implicit Attribute Value Extraction (AVE) is essential for accurately representing products in e-commerce due to the complexity of multidimensional data and gaps in vision-text understanding.
Approach:	They propose a multi-agent framework that employs multiple MLLM agents to iteratively refine inferences through debate rounds.
Outcome:	The proposed framework improves inference performance and robustness through debate rounds.

Data Augmentation for Low-Resource Keyphrase Generation (2023.findings-acl)

Copied to clipboard

Krishna Garg, Jishnu Ray Chowdhury, Cornelia Caragea

The Myth of Double-Blind Review Revisited: ACL vs. EMNLP (D19-1)

Challenge:	Existing works on keyphrase generation rely on large-scale annotated datasets, which are not easy to acquire.
Approach:	They propose to use full text to improve keyphrase generation in resource-constrained domains by using the full text of the articles to augment their methods.
Outcome:	The proposed methods improve both present and absent keyphrase generation on three datasets and show that they are cost-effective.

Copied to clipboard

Cornelia Caragea, Ana Uban, Liviu P. Dinu

SciNLI: A Corpus for Natural Language Inference on Scientific Text (2022.acl-long)

Challenge:	a double-blind review system enforces author anonymity during the review period . authors can be inferred with accuracy as high as 87% on ACL and 78% on EMNLP .
Approach:	They examine how well deep learning techniques can infer authors of a paper . authors found authors can be inferred with accuracy as high as 87% on ACL and 78% on EMNLP .
Outcome:	The proposed method can infer authors with 87% accuracy on ACL and 78% on EMNLP for the top 100 most prolific authors.

Copied to clipboard

A MISMATCHED Benchmark for Scientific Natural Language Inference (2025.findings-acl)

Reasoning

Challenge:	Existing Natural Language Inference (NLI) datasets are not related to scientific text.
Approach:	They propose a large dataset for NLI that captures the formality in scientific text and contains 107,412 sentence pairs extracted from scholarly papers on NLP and computational linguistics.
Outcome:	The proposed model achieves a Macro F1 score of only 78.18% and an accuracy of 78.23%.

Copied to clipboard

Firoz Shaik, Mobashir Sadat, Nikita Gautam, Doina Caragea, Cornelia Caragea

C-STANCE: A Large Dataset for Chinese Zero-Shot Stance Detection (2023.acl-long)

Challenge:	Existing datasets for scientific NLI are derived from various computer science domains, whereas non-CS domains are completely ignored.
Approach:	They propose a scientific natural language inference benchmark called MisMatched that incorporates sentence pairs having an implicit scientific NLI relation into model training.
Outcome:	The proposed benchmark covers three non-CS domains and contains 2,700 human annotated sentence pairs.

Copied to clipboard

Chenye Zhao, Yingjie Li, Cornelia Caragea

DeCrisisMB: Debiased Semi-Supervised Learning for Crisis Tweet Classification via Memory Bank (2023.findings-emnlp)

Challenge:	Recent advances in zero-shot stance detection are limited to English and Chinese . stance can provide useful information for important events such as policymaking and presidential elections.
Approach:	They present a Chinese dataset for zero-shot stance detection that is the first for ZSSD.
Outcome:	The proposed dataset is the first Chinese dataset for zero-shot stance detection.

Copied to clipboard

Henry Zou, Yue Zhou, Weizhi Zhang, Cornelia Caragea

Keyphrase Generation Beyond the Boundaries of Title and Abstract (2022.findings-emnlp)

Challenge:	Existing studies utilize social media platforms such as Twitter to build models for crisis event analysis, but semi-supervised approaches require annotating vast amounts of data and are impractical due to limited response time.
Approach:	They propose a method that stores and performs equal sampling for generated pseudo-labels from each class at each training iteration.
Outcome:	The proposed method performs better than existing methods in both in-distribution and out-of-difference settings.

Copied to clipboard

Krishna Garg, Jishnu Ray Chowdhury, Cornelia Caragea

Text Generation

Challenge:	Current approaches to keyphrase generation use only the title and abstract of the articles.
Approach:	They propose to integrate full text and semantically similar articles to generate keyphrases from a dataset that includes the full text of the articles along with the title and abstract.
Outcome:	The proposed model can generate keyphrases that are present or absent from the text.

Co-training for Low Resource Scientific Natural Language Inference (2024.acl-long)

Copied to clipboard

Sarcasm Detection in a Disaster Context (2024.lrec-main)

Challenge:	Scientific Natural Language Inference (NLI) is a task to predict the semantic relation between sentences extracted from research articles.
Approach:	They propose a co-training method that assigns weights based on the training dynamics of the classifiers to the distantly supervised labels.
Outcome:	The proposed method achieves an improvement of 1.5% in Macro F1 over the distant supervision baseline and substantial improvements over several other strong SSL baselines.

Copied to clipboard

Tiberiu Sosea, Junyi Jessy Li, Cornelia Caragea

Language Models (Mostly) Do Not Consider Emotion Triggers When Predicting Emotion (2024.naacl-short)

Challenge:	During natural disasters, people often use social media platforms to express contempt or sarcasm . despite being widely researched as an NLP task, sarkasmatic detection has not been explored in a specific context .
Approach:	They propose a dataset of 15,000 tweets annotated for intended sarcasm . they propose sarkasmatic detection using pre-trained language models .
Outcome:	The proposed model can obtain as much as 0.70 F1 on the dataset.

Copied to clipboard

Smriti Singh, Cornelia Caragea, Junyi Jessy Li

Detecting Perceived Emotions in Hurricane Disasters (2020.acl-main)

Challenge:	Existing work has sought to identify what triggers or causes a particular emotion, but the relationship between those triggers and the prediction of emotion detection models is little understood.
Approach:	They propose a dataset to evaluate the ability of large language models to identify emotion triggers . they compare features considered important for emotion prediction models to those considered less salient .
Outcome:	The proposed dataset compares large language models and fine-tuned models on social media posts . it shows that emotion triggers are not considered salient features for emotion prediction models .

Copied to clipboard

Shrey Desai, Cornelia Caragea, Junyi Jessy Li

Detecting Optimism in Tweets using Knowledge Distillation and Linguistic Analysis of Optimism (2022.lrec-1)

Challenge:	Existing methods for emotion detection are limited in disaster-centric domains due to distributional shifts.
Approach:	They propose to use a Twitter emotion dataset to analyze emotions in natural disasters . they propose to apply classification tasks to discriminate between coarse-grained emotions .
Outcome:	The proposed model achieves only 68% accuracy after pre-training with unlabeled Twitter data.

Copied to clipboard

Ștefan Cobeli, Ioan-Bogdan Iordache, Shweta Yadav, Cornelia Caragea, Liviu P. Dinu, Dragoș Iliescu

Identifying Medical Self-Disclosure in Online Communities (2021.naacl-main)

Challenge:	a recent study has established sentiment analysis as an alluring problem, but many feelings are left unexplored.
Approach:	They propose a framework to learn the polarity of emotions from Twitter posts . they compare optimism detection with sentiment analysis and hate speech detection .
Outcome:	The proposed framework differs between optimistic and pessimistic users on the Optimism/Pessimism Twitter dataset.

Copied to clipboard

Mina Valizadeh, Pardis Ranjbar-Noiey, Cornelia Caragea, Natalie Parde

EZ-STANCE: A Large Dataset for Zero-Shot Stance Detection (2023.findings-emnlp)

Challenge:	a new dataset of health-related posts from online social platforms is available for analysis . medical self-disclosure may be useful for early detection and treatment of medical issues .
Approach:	They propose to analyze medical self-disclosure in online health conversations . they release a dataset of health-related posts from online social platforms with high inter-annotator agreement .
Outcome:	The proposed model achieves an accuracy of 81.02% and sets a strong performance benchmark.

Copied to clipboard

Chenye Zhao, Cornelia Caragea

Bilingual Zero-Shot Stance Detection (2025.acl-long)

Challenge:	EZ-STANCE is a large dataset for zero-shot stance detection in english . it includes both noun-phrase targets and claim targets covering a wide range of domains.
Approach:	They present a large English ZSSD dataset with 30,606 annotated text-target pairs . they propose to transform EZ-STANCE into the NLI task by applying two simple yet effective prompts to noun-phrase targets.
Outcome:	The proposed dataset includes noun-phrase targets and claim targets covering a wide range of domains.

Copied to clipboard

Chenye Zhao, Cornelia Caragea

MSciNLI: A Diverse Benchmark for Scientific Natural Language Inference (2024.naacl-long)

Challenge:	a study focuses on noun-phrase and claim targets within bilingual ZSSD scenarios . a dataset focusing on claim targets with a low occurrence of shared words is also explored .
Approach:	They use a bilingual bilingual ZSSD dataset to investigate the use of zero-shot stance detection.
Outcome:	The proposed dataset is the first to examine this difficult setting in bilingual ZSSD . it focuses on noun-phrase and claim targets within in-domain and out-of-domain bilingual scenarios .

Copied to clipboard

Model Calibration for Emotion Detection (2025.findings-emnlp)

Challenge:	a dataset containing 132,320 sentence pairs from five new scientific domains is used for scientific Natural Language Inference (NLI) the availability of multiple domains makes it possible to study domain shift for scientific NLI.
Approach:	They propose a dataset with 132,320 sentence pairs from five new scientific domains to introduce diversity in scientific NLI.
Outcome:	The proposed dataset contains 132,320 sentence pairs extracted from five new scientific domains.

Copied to clipboard

Mihaela Petre-Vlad, Cornelia Caragea, Florentina Hristea

Exploiting Position and Contextual Word Embeddings for Keyphrase Extraction from Scientific Papers (2021.eacl-main)

Challenge:	a MixUp method is used to calibrate emotion detection models based on knowledge distillation and the MixUp data augmentation technique.
Approach:	They propose a method that uses knowledge distillation and the MixUp data augmentation technique to enhance the trustworthiness of emotion detection models.
Outcome:	The proposed method improves the accuracy of the teacher models and the student models.

Copied to clipboard

Krutarth Patel, Cornelia Caragea

Stanceformer: Target-Aware Transformer for Stance Detection (2024.findings-emnlp)

Challenge:	Existing methods for keyphrase extraction are either supervised or unsupervised.
Approach:	They propose an unsupervised algorithm that exploits contextual word embeddings and positional information to create a biased PageRank.
Outcome:	The proposed algorithm outperforms previous approaches and strong baselines on five benchmark datasets.

Copied to clipboard

Krishna Garg, Cornelia Caragea

Emotion analysis and detection during COVID-19 (2022.lrec-1)

Challenge:	Existing transformer models that lack the capability to prioritize targets under-perform and are underperforming the task.
Approach:	They propose a target-aware transformer model that incorporates enhanced attention towards the targets during both training and inference.
Outcome:	The proposed model improves on state-of-the-art models and Large Language Models and can be used for other domains.

Copied to clipboard

Tiberiu Sosea, Chau Pham, Alexander Tekle, Cornelia Caragea, Junyi Jessy Li

P-Stance: A Large Dataset for Stance Detection in Political Domain (2021.findings-acl)

Challenge:	3,000 English tweets labeled with emotions are used to predict emotions during crises . authors propose semi-supervised learning to bridge this gap .
Approach:	They propose to use a dataset of 3,000 English tweets labeled with emotions . they propose semi-supervised learning to bridge this gap by analyzing unlabeled data .
Outcome:	The proposed model can be used to predict emotions in the context of COVID-19 . the proposed model performs better than other models using unlabeled data .

Copied to clipboard

Yingjie Li, Tiberiu Sosea, Aditya Sawant, Ajith Jayaraman Nair, Diana Inkpen, Cornelia Caragea

A Novel Cartography-Based Curriculum Learning Method Applied on RoNLI: The First Romanian Natural Language Inference Corpus (2024.acl-long)

Challenge:	stance detection is a method to determine whether a text author is in favor of, against or neutral toward a specific target.
Approach:	They propose to use a large stance detection dataset in the political domain to detect stances on twitter.
Outcome:	The proposed model achieves a macro-average F1-score of 80.53% and can be used to improve cross-domain stance detection.

Copied to clipboard

Eduard Poesina, Cornelia Caragea, Radu Ionescu

SciDMT: A Large-Scale Corpus for Detecting Scientific Mentions (2024.lrec-main)

Challenge:	Natural language inference (NLI) is an actively studied topic serving as a proxy for natural language understanding.
Approach:	They propose to use a Romanian NLI corpus to analyze sentence pairs . they use multiple machine learning methods to establish competitive baselines .
Outcome:	The proposed model improves on the best model by employing a new curriculum learning strategy based on data cartography.

Copied to clipboard

Huitong Pan, Qi Zhang, Cornelia Caragea, Eduard Dragut, Longin Jan Latecki

Challenge:	SciDMT is an enhanced and expanded corpus for scientific mention detection . existing corpora are limited by their small volume and entity linking capabilities .
Approach:	They propose to enhance SciDMT, an annotated scientific corpus for scientific mention detection.
Outcome:	The proposed corpus is the largest for scientific entity mention detection . it is based on deep learning architectures like SciBERT and GPT-3.5 .

GunStance: Stance Detection for Gun Control and Gun Regulation (2024.acl-long)

Copied to clipboard

Nikesh Gyawali, Iustin Sirbu, Tiberiu Sosea, Sarthak Khanal, Doina Caragea, Traian Rebedea, Cornelia Caragea

DMDD: A Large-Scale Dataset for Dataset Mentions Detection (2023.tacl-1)

Challenge:	Social media, especially Twitter, has been a melting pot for such debates.
Approach:	They propose to annotate tweets relevant to shooting events into three classes: In-Favor, Against, and Neutral.
Outcome:	The proposed approach outperforms supervised, semi-supervised, and LLM-based zero-shot models on the dataset.

Copied to clipboard

Huitong Pan, Qi Zhang, Eduard Dragut, Cornelia Caragea, Longin Jan Latecki

ImplicitAVE: An Open-Source Dataset and Multimodal LLMs Benchmark for Implicit Attribute Value Extraction (2024.findings-acl)

Challenge:	Existing corpora for dataset mention detection are limited in size and naming diversity.
Approach:	They propose a dataset for dataset mention detection that is the largest publicly available corpus for this task.
Outcome:	The proposed dataset is the largest publicly available corpus for dataset mention detection . it identifies open problems in dataset mention recognition and linking .

Copied to clipboard

Henry Zou, Vinay Samuel, Yue Zhou, Weizhi Zhang, Liancheng Fang, Zihe Song, Philip Yu, Cornelia Caragea

Studying the Evolution of Scientific Topics and their Relationships (2021.findings-acl)

Challenge:	Existing datasets for attribute value extraction focus on explicit attribute values while neglecting the implicit ones.
Approach:	They present a multimodal dataset for implicit attribute value extraction that includes AVE and multimodality.
Outcome:	The proposed dataset includes 68k training and 1.6k testing data across five domains.

Copied to clipboard

Ana Sabina Uban, Cornelia Caragea, Liviu P. Dinu

A New Direction in Stance Detection: Target-Stance Extraction in the Wild (2023.acl-long)

Challenge:	a study of scientific topics and their evolution through time is proposed . we analyze scientific texts published in the field of computational linguistics .
Approach:	They propose a multidimensional approach to studying scientific topics through time and their relationships between them.
Outcome:	The proposed model analyzes scientific texts published in the ACL Anthology and compares them with case studies to understand how topics evolve and disappear over time.

Copied to clipboard

Yingjie Li, Krishna Garg, Cornelia Caragea

Distilling Knowledge for Empathy Detection (2021.findings-emnlp)

Challenge:	Existing methods for stance detection assume that the target is known in advance . Existing tasks use implicit mentions in the source text and are infeasible to have manual annotations at a large scale.
Approach:	They propose a task Target-Stance Extraction that aims to extract the (target, stance) pair from social media texts.
Outcome:	The proposed task can facilitate future research in the field of stance detection.

Copied to clipboard

Mahshid Hosseini, Cornelia Caragea

Multi-Task Knowledge Distillation with Embedding Constraints for Scholarly Keyphrase Boundary Classification (2023.emnlp-main)

Challenge:	Empathy is the link between self and others.
Approach:	They employ multi-task training with knowledge distillation to integrate knowledge from available resources to detect empathy from the natural language in different domains.
Outcome:	The proposed approach yields better results on an existing news-related empathy dataset compared to strong baselines.

Copied to clipboard

Seo Park, Cornelia Caragea

Active Few-Shot Learning for Text Classification (2025.naacl-long)

Challenge:	Scholarly keyphrase boundary classification is the task of identifying highly summative phrases from scientific papers and classifying them into a set of predefined classes.
Approach:	They propose a constraint which enforces the teachers and student similarity in the embedding space.
Outcome:	The proposed constraint outperforms previous studies and strong baselines on three datasets of scientific documents.

Copied to clipboard

Saeed Ahmadnia, Arash Yousefi Jordehi, Mahsa Hosseini Khasheh Heyran, Seyed Abolghasem Mirroshandel, Owen Rambow, Cornelia Caragea

CancerEmo: A Dataset for Fine-Grained Emotion Detection (2020.emnlp-main)

Challenge:	Recent advances in Large Language Models (LLMs) have boosted the use of Few-Shot Learning (FSL) methods in natural language processing.
Approach:	They propose a method that identifies effective support instances from the unlabeled pool and can work with different LLMs.
Outcome:	The proposed method improves on five tasks on which it is tested on five LLMs.

Copied to clipboard

Scientific Keyphrase Identification and Classification by Pre-Trained Language Models Intermediate Task Transfer Learning (2020.coling-main)

Challenge:	a lack of large annotated datasets hinders emotion detection in the health domain . a recent study shows that online sharing of emotions is beneficial to a patient's progress .
Approach:	They propose an emotion dataset annotated with eight fine-grained emotions from an online health community.
Outcome:	The proposed model achieves an average F1 of 71% on the cancerEmo dataset . the best model achieve a higher F1 than the previous model, which was improved using domain-specific pre-training.

Copied to clipboard

Seoyeon Park, Cornelia Caragea

Why Do You Feel This Way? Summarizing Triggers of Emotions in Social Media Posts (2022.emnlp-main)

Challenge:	Scientific keyphrase identification and classification is challenging due to lack of labeled data for training deep neural networks.
Approach:	They propose to use pre-trained language models BERT and SciBERT to train scientific keyphrase identification and classification using intermediate task transfer learning.
Outcome:	The proposed model achieves competitive performance in scientific keyphrase identification and classification compared to both previous studies and strong baselines.

Copied to clipboard

Hongli Zhan, Tiberiu Sosea, Cornelia Caragea, Junyi Jessy Li

Text Generation

Challenge:	Large-scale crises such as the COVID-19 pandemic cause emotional turmoil worldwide.
Approach:	They propose a method to jointly detect emotions and summarize emotion triggers in social media posts related to COVID-19.
Outcome:	The proposed method can detect emotions and summarize emotions in long social media posts.

Fine-Grained Emotion Detection in Health-Related Online Posts (D18-1)

Copied to clipboard

Hamed Khanpour, Cornelia Caragea

Learning to Infer from Unlabeled Data: A Semi-supervised Learning Approach for Robust Natural Language Inference (2022.findings-emnlp)

Challenge:	Emotion detection from health-related posts is based on a health-specific vocabulary that people use in OHCs.
Approach:	They propose to use deep neural networks and lexicon-based features to detect emotions in health-related posts.
Outcome:	The proposed method uses high-level and abstract features derived from deep neural networks combined with lexicon-based features to detect emotions.

Copied to clipboard

What Clued the AI Doctor In? On the Influence of Data Source and Quality for Transformer-Based Medical Self-Disclosure Detection (2023.eacl-main)

Challenge:	Semi-supervised learning (SSL) is a popular technique for reducing the reliance on human annotations for NLI tasks.
Approach:	They propose a way to incorporate unlabeled data into semi-supervised learning (SSL) using a conditional language model, they propose to generate hypotheses for unlabed sentences .
Outcome:	The proposed framework significantly improves the performance of four NLI datasets in low-resource settings.

Copied to clipboard

Mina Valizadeh, Xing Qian, Pardis Ranjbar-Noiey, Cornelia Caragea, Natalie Parde

Calibrating Student Models for Emotion-related Tasks (2022.emnlp-main)

Challenge:	Recognizing medical self-disclosure is important in many healthcare contexts, but it has been under-explored by the NLP community.
Approach:	They analyze a social media-based task to expand existing medical self-disclosure corpus and compare Transformer-based models to determine their merits.
Outcome:	The proposed dataset outperforms the state-of-the-art dataset for this task by 16.73%.

Copied to clipboard

Mahshid Hosseini, Cornelia Caragea

LLM-Guided Co-Training for Text Classification (2025.emnlp-main)

Challenge:	Knowledge distillation is an effective method to transfer knowledge from one network (a.k.a. teacher) to another (as student).
Approach:	They propose to use a mixup data augmentation technique to increase the accuracy of the model by providing better training signals to the student models.
Outcome:	The proposed method improves the calibration of student models while providing better training signals to the student models using training dynamics.

Copied to clipboard

Md Mezbaur Rahman, Cornelia Caragea