arXiv subjects

Doina Caragea

Publications and source records attributed to Doina Caragea.

At least 19 recordsLinked to original sources

arXiv2026

Multi-Agent Reasoning with Adaptive Worker Allocation for Stance Detection

Stance detection requires identifying an author's position toward a target, often from short-form texts where stance is implicit, indirect, or rhetorically framed. Although large language models (LLMs) achieve strong performance on this task, single-pass prompting can be brittle when multiple interpretations are plausible. Existing aggregation strategies, such as majority voting or self-consistency, improve robustness by combining labels, but they discard the intermediate reasoning needed to resolve conflicting interpretations. We introduce a multi-agent reasoning framework with adaptive worker allocation for stance detection that shifts aggregation from label-level voting to reasoning-level synthesis. The framework employs a Manager-Worker architecture in which a Manager adaptively allocates a variable number of Worker agents based on input complexity. Each Worker analyzes the input from a distinct perspective and produces a reasoning-only explanation without emitting a stance label; the Manager then synthesizes these explanations to produce the final prediction. We evaluate the proposed framework on SemEval-2016, P-Stance, and COVID-19 Stance using Llama, Mistral, and Gemini. Results show that the framework yields the largest gains on implicit and context-dependent stance cases, achieving 86.07 Macro-F1 on COVID-19 and 82.90 on SemEval-2016, while remaining competitive on more explicit stance datasets such as P-Stance. These findings suggest that adaptive reasoning-level aggregation is most beneficial when stance cannot be reliably inferred from surface cues alone.

cs.CL↗

arXiv2026

LLM-guided Semi-Supervised Approaches for Social Media Crisis Data Classification

Semi-supervised learning approaches have been investigated as a means to enhance the analysis of social media data in disaster management contexts. In this work, we present the first empirical evaluation of large language model (LLM) guided semi-supervised learning for crisis related tweet classification. We compare two recent LLM assisted semi-supervised methods, VerifyMatch and LLM guided Co-Training ( LG-CoTrain), against established semi-supervised baselines. Our results show that LG-CoTrain significantly outperforms classical semi-supervised approaches in low resource settings with 5, 10 and 25 labeled examples per class, achieving the highest averaged Macro F1 across events. VerifyMatch achieves competitive performance while also demonstrating strong calibration properties. As the number of labeled examples increases, the performance gap narrows and Self Training emerges as a strong baseline. We further observe that compact semi-supervised models can, in some cases, outperform very large LLMs operating in zero-shot settings. This finding highlights the potential of transferring knowledge from LLMs into smaller and more deployable models through LLM guided semi-supervised learning, offering a practical pathway for real world disaster response applications. Our project repository on Github is here.

cs.AI↗

arXiv2026

MSGL-Transformer: A Multi-Scale Global-Local Transformer for Rodent Social Behavior Recognition

Recognition of rodent behavior is important for understanding neural and behavioral mechanisms. Traditional manual scoring is time-consuming and prone to human error. We propose MSGL-Transformer, a Multi-Scale Global-Local Transformer for recognizing rodent social behaviors from pose-based temporal sequences. The model employs a lightweight transformer encoder with multi-scale attention to capture motion dynamics across different temporal scales. The architecture integrates parallel short-range, medium-range, and global attention branches to explicitly capture behavior dynamics at multiple temporal scales. We also introduce a Behavior-Aware Modulation (BAM) block, inspired by SE-Networks, which modulates temporal embeddings to emphasize behavior-relevant features prior to attention. We evaluate on two datasets: RatSI (5 behavior classes, 12D pose inputs) and CalMS21 (4 behavior classes, 28D pose inputs). On RatSI, MSGL-Transformer achieves 75.4% mean accuracy and F1-score of 0.745 across nine cross-validation splits, outperforming TCN, LSTM, and Bi-LSTM. On CalMS21, it achieves 87.1% accuracy and F1-score of 0.8745, a +10.7% improvement over HSTWFormer, and outperforms ST-GCN, MS-G3D, CTR-GCN, and STGAT. The same architecture generalizes across both datasets with only input dimensionality and number of classes adjusted.

cs.CV↗

arXiv2026

Leveraging Large Language Models for Automated Scalable Development of Open Scientific Databases

With the exponential increase in online scientific literature, identifying reliable domain-specific data has become increasingly important but also very challenging. Manual data collection and filtering for domain-specific scientific literature is not only time-consuming but also labor-intensive and prone to errors and inconsistencies. To facilitate automated data collection, the paper introduces a web-based tool that leverages Large Language Models (LLMs) for automated and scalable development of open scientific databases. More specifically, the tool is based on an automated and unified framework that combines keyword-based querying, API-enabled data retrieval, and LLM-powered text classification to construct domain-specific scientific databases. Data is collected from multiple reliable data sources and search engines using a parallel querying technique to construct a combined unified dataset. The dataset is subsequently filtered using LLMs queried with prompts tailored for each keyword-based query to extract the relevant data to a scientific query of interest. The approach was tested across a set of variable keyword-based searches for different domain-specific tasks related to agriculture and crop yield. The results and analysis show 90\% overlap with small domain expert-curated databases, suggesting that the proposed tool can be used to significantly reduce manual workload. Furthermore, the proposed framework is both scalable and domain-agnostic and can be applied across diverse fields for building scalable open scientific databases.

cs.IR↗

arXiv2026

Neural Network Based Molecular Structure Retrieval from Coulomb Explosion Imaging Data

Determining the structure and following the structural evolution of molecules undergoing chemical reactions is one of the key goals of ultrafast molecular physics and chemistry. Recently, Coulomb explosion imaging has emerged as a promising technique for imaging the evolving structure of individual molecules in the gas phase. However, its practical application for structure determination is hampered by the lack of suitable algorithms to directly retrieve the molecular structure from the measured fragment-ion momentum data. Here, we propose a scheme to solve the underlying inverse problem by employing neural networks to infer the initial atomic positions from the final ion momenta on an event-by-event basis. Using this scheme, we retrieve the structure of several polyhalomethane isomers from simulated Coulomb explosion imaging data with an average per-atom position error of approximately 0.1 atomic units, i.e., to within 5% of the typical bond lengths. This development paves the way for an automated structure retrieval from Coulomb explosion data one molecule at a time, making it ideally suitable for analyzing pump-probe experiments where several products are formed that need to be distinguished.

physics.chem-ph↗

arXiv2026

Practical Insights into Semi-Supervised Object Detection Approaches

Learning in data-scarce settings has recently gained significant attention in the research community. Semi-supervised object detection(SSOD) aims to improve detection performance by leveraging a large number of unlabeled images alongside a limited number of labeled images(a.k.a.,few-shot learning). In this paper, we present a comprehensive comparison of three state-of-the-art SSOD approaches, including MixPL, Semi-DETR and Consistent-Teacher, with the goal of understanding how performance varies with the number of labeled images. We conduct experiments using the MS-COCO and Pascal VOC datasets, two popular object detection benchmarks which allow for standardized evaluation. In addition, we evaluate the SSOD approaches on a custom Beetle dataset which enables us to gain insights into their performance on specialized datasets with a smaller number of object categories. Our findings highlight the trade-offs between accuracy, model size, and latency, providing insights into which methods are best suited for low-data regimes.

cs.CV↗

arXiv2025

The Shifting Landscape of Vaccine Discourse: Insights From a Decade of Pre- to Post-COVID-19 Vaccine Posts on Social Media

In this work, we study English-language vaccine discourse in social media posts, specifically posts on X (formerly Twitter), in seven years before the COVID-19 outbreak (2013 to 2019) and three years after the outbreak was first reported (2020 to 2022). Drawing on theories from social cognition and the stereotype content model in Social Psychology, we analyze how English speakers talk about vaccines on social media to understand the evolving narrative around vaccines in social media posts. To do that, we first introduce a novel dataset comprising 18.7 million curated posts on vaccine discourse from 2013 to 2022. This extensive collection-filtered down from an initial 129 million posts through rigorous preprocessing-captures both pre-COVID and COVID-19 periods, offering valuable insights into the evolution of English-speaking X users' perceptions related to vaccines. Our analysis shows that the COVID-19 pandemic led to complex shifts in X users' sentiment and discourse around vaccines. We observe that negative emotion word usage decreased during the pandemic, with notable rises in usage of surprise, and trust related emotion words. Furthermore, vaccine-related language tended to use more warmth-focused words associated with trustworthiness, along with positive, competence-focused words during the early days of the pandemic, with a marked rise in negative word usage towards the end of the pandemic, possibly reflecting a growing vaccine hesitancy and skepticism.

cs.SI↗

arXiv2025

Evaluating Large Language Models for Stance Detection on Financial Targets from SEC Filing Reports and Earnings Call Transcripts

Financial narratives from U.S. Securities and Exchange Commission (SEC) filing reports and quarterly earnings call transcripts (ECTs) are very important for investors, auditors, and regulators. However, their length, financial jargon, and nuanced language make fine-grained analysis difficult. Prior sentiment analysis in the financial domain required a large, expensive labeled dataset, making the sentence-level stance towards specific financial targets challenging. In this work, we introduce a sentence-level corpus for stance detection focused on three core financial metrics: debt, earnings per share (EPS), and sales. The sentences were extracted from Form 10-K annual reports and ECTs, and labeled for stance (positive, negative, neutral) using the advanced ChatGPT-o3-pro model under rigorous human validation. Using this corpus, we conduct a systematic evaluation of modern large language models (LLMs) using zero-shot, few-shot, and Chain-of-Thought (CoT) prompting strategies. Our results show that few-shot with CoT prompting performs best compared to supervised baselines, and LLMs' performance varies across the SEC and ECT datasets. Our findings highlight the practical viability of leveraging LLMs for target-specific stance in the financial domain without requiring extensive labeled data.

cs.CL↗

arXiv2025

Predictive Modeling and Explainable AI for Veterinary Safety Profiles, Residue Assessment, and Health Outcomes Using Real-World Data and Physicochemical Properties

The safe use of pharmaceuticals in food-producing animals is vital to protect animal welfare and human food safety. Adverse events (AEs) may signal unexpected pharmacokinetic or toxicokinetic effects, increasing the risk of violative residues in the food chain. This study introduces a predictive framework for classifying outcomes (Death vs. Recovery) using ~1.28 million reports (1987-2025 Q1) from the U.S. FDA's OpenFDA Center for Veterinary Medicine. A preprocessing pipeline merged relational tables and standardized AEs through VeDDRA ontologies. Data were normalized, missing values imputed, and high-cardinality features reduced; physicochemical drug properties were integrated to capture chemical-residue links. We evaluated supervised models, including Random Forest, CatBoost, XGBoost, ExcelFormer, and large language models (Gemma 3-27B, Phi 3-12B). Class imbalance was addressed, such as undersampling and oversampling, with a focus on prioritizing recall for fatal outcomes. Ensemble methods(Voting, Stacking) and CatBoost performed best, achieving precision, recall, and F1-scores of 0.95. Incorporating Average Uncertainty Margin (AUM)-based pseudo-labeling of uncertain cases improved minority-class detection, particularly in ExcelFormer and XGBoost. Interpretability via SHAP identified biologically plausible predictors, including lung, heart, and bronchial disorders, animal demographics, and drug physicochemical properties. These features were strongly linked to fatal outcomes. Overall, the framework shows that combining rigorous data engineering, advanced machine learning, and explainable AI enables accurate, interpretable predictions of veterinary safety outcomes. The approach supports FARAD's mission by enabling early detection of high-risk drug-event profiles, strengthening residue risk assessment, and informing regulatory and clinical decision-making.

cs.LG↗

arXiv2025

AutoPK: Leveraging LLMs and a Hybrid Similarity Metric for Advanced Retrieval of Pharmacokinetic Data from Complex Tables and Documents

Pharmacokinetics (PK) plays a critical role in drug development and regulatory decision-making for human and veterinary medicine, directly affecting public health through drug safety and efficacy assessments. However, PK data are often embedded in complex, heterogeneous tables with variable structures and inconsistent terminologies, posing significant challenges for automated PK data retrieval and standardization. AutoPK, a novel two-stage framework for accurate and scalable extraction of PK data from complex scientific tables. In the first stage, AutoPK identifies and extracts PK parameter variants using large language models (LLMs), a hybrid similarity metric, and LLM-based validation. The second stage filters relevant rows, converts the table into a key-value text format, and uses an LLM to reconstruct a standardized table. Evaluated on a real-world dataset of 605 PK tables, including captions and footnotes, AutoPK shows significant improvements in precision and recall over direct LLM baselines. For instance, AutoPK with LLaMA 3.1-70B achieved an F1-score of 0.92 on half-life and 0.91 on clearance parameters, outperforming direct use of LLaMA 3.1-70B by margins of 0.10 and 0.21, respectively. Smaller models such as Gemma 3-27B and Phi 3-12B with AutoPK achieved 2-7 fold F1 gains over their direct use, with Gemma's hallucination rates reduced from 60-95% down to 8-14%. Notably, AutoPK enabled open-source models like Gemma 3-27B to outperform commercial systems such as GPT-4o Mini on several PK parameters. AutoPK enables scalable and high-confidence PK data extraction, making it well-suited for critical applications in veterinary pharmacology, drug safety monitoring, and public health decision-making, while addressing heterogeneous table structures and terminology and demonstrating generalizability across key PK parameters. Code and data: https://github.com/hosseinsholehrasa/AutoPK

cs.DB↗

arXiv2025

Benchmarking Android Malware Detection: Traditional vs. Deep Learning Models

Android malware detection has been extensively studied using both traditional machine learning (ML) and deep learning (DL) approaches. While many state-of-the-art detection models, particularly those based on DL, claim superior performance, they often rely on limited comparisons, lacking comprehensive benchmarking against traditional ML models across diverse datasets. This raises concerns about the robustness of DL-based approaches' performance and the potential oversight of simpler, more efficient ML models. In this paper, we conduct a systematic evaluation of Android malware detection models across four datasets: three recently published, publicly available datasets and a large-scale dataset we systematically collected. We implement a range of traditional ML models, including Random Forests (RF) and CatBoost, alongside advanced DL models such as Capsule Graph Neural Networks (CapsGNN), BERT-based models, and ExcelFormer based models. Our results reveal that in many cases simpler and more computationally efficient ML models achieve comparable or even superior performance compared with DL models. These findings highlight the need for rigorous benchmarking in Android malware detection research. We encourage future studies to conduct more comprehensive benchmarking comparisons between traditional and advanced models to ensure a more accurate assessment of detection capabilities. To facilitate further research, we provide access to our dataset, including app IDs, hash values, and labels.

cs.CR↗

arXiv2025

The Impact of Train-Test Leakage on Machine Learning-based Android Malware Detection

When machine learning is used for Android malware detection, an app needs to be represented in a numerical format for training and testing. We identify a widespread occurrence of distinct Android apps that have identical or nearly identical app representations. In particular, among app samples in the testing dataset, there can be a significant percentage of apps that have an identical or nearly identical representation to an app in the training dataset. This will lead to a data leakage problem that inflates a machine learning model's performance as measured on the testing dataset. The data leakage not only could lead to overly optimistic perceptions on the machine learning models' ability to generalize beyond the data on which they are trained, in some cases it could also lead to qualitatively different conclusions being drawn from the research. We present two case studies to illustrate this impact. In the first case study, the data leakage inflated the performance results but did not impact the overall conclusions made by the researchers in a qualitative way. In the second case study, the data leakage problem would have led to qualitatively different conclusions being drawn from the research. We further examine the real-world impact of the data leakage by dissecting the capability of memorization and the capability of generalization of a machine learning model, and show that by removing leakage from testing data, the evaluation results better reflect the machine learning model's utility in real-world Android malware detection scenarios.

cs.CR↗

arXiv2025

A MISMATCHED Benchmark for Scientific Natural Language Inference

Scientific Natural Language Inference (NLI) is the task of predicting the semantic relation between a pair of sentences extracted from research articles. Existing datasets for this task are derived from various computer science (CS) domains, whereas non-CS domains are completely ignored. In this paper, we introduce a novel evaluation benchmark for scientific NLI, called MISMATCHED. The new MISMATCHED benchmark covers three non-CS domains-PSYCHOLOGY, ENGINEERING, and PUBLIC HEALTH, and contains 2,700 human annotated sentence pairs. We establish strong baselines on MISMATCHED using both Pre-trained Small Language Models (SLMs) and Large Language Models (LLMs). Our best performing baseline shows a Macro F1 of only 78.17% illustrating the substantial headroom for future improvements. In addition to introducing the MISMATCHED benchmark, we show that incorporating sentence pairs having an implicit scientific NLI relation between them in model training improves their performance on scientific NLI. We make our dataset and code publicly available on GitHub.

cs.CL↗

arXiv2023

CrisisMatch: Semi-Supervised Few-Shot Learning for Fine-Grained Disaster Tweet Classification

The shared real-time information about natural disasters on social media platforms like Twitter and Facebook plays a critical role in informing volunteers, emergency managers, and response organizations. However, supervised learning models for monitoring disaster events require large amounts of annotated data, making them unrealistic for real-time use in disaster events. To address this challenge, we present a fine-grained disaster tweet classification model under the semi-supervised, few-shot learning setting where only a small number of annotated data is required. Our model, CrisisMatch, effectively classifies tweets into fine-grained classes of interest using few labeled data and large amounts of unlabeled data, mimicking the early stage of a disaster. Through integrating effective semi-supervised learning ideas and incorporating TextMixUp, CrisisMatch achieves performance improvement on two disaster datasets of 11.2\% on average. Further analyses are also provided for the influence of the number of labeled data and out-of-domain results.

cs.CL↗

arXiv2021

Identification of Fine-Grained Location Mentions in Crisis Tweets

Identification of fine-grained location mentions in crisis tweets is central in transforming situational awareness information extracted from social media into actionable information. Most prior works have focused on identifying generic locations, without considering their specific types. To facilitate progress on the fine-grained location identification task, we assemble two tweet crisis datasets and manually annotate them with specific location types. The first dataset contains tweets from a mixed set of crisis events, while the second dataset contains tweets from the global COVID-19 pandemic. We investigate the performance of state-of-the-art deep learning models for sequence tagging on these datasets, in both in-domain and cross-domain settings.

cs.CL↗

arXiv2020

On Identifying Hashtags in Disaster Twitter Data

Tweet hashtags have the potential to improve the search for information during disaster events. However, there is a large number of disaster-related tweets that do not have any user-provided hashtags. Moreover, only a small number of tweets that contain actionable hashtags are useful for disaster response. To facilitate progress on automatic identification (or extraction) of disaster hashtags for Twitter data, we construct a unique dataset of disaster-related tweets annotated with hashtags useful for filtering actionable information. Using this dataset, we further investigate Long Short Term Memory-based models within a Multi-Task Learning framework. The best performing model achieves an F1-score as high as 92.22%. The dataset, code, and other resources are available on Github.

cs.IR↗

arXiv2019

Keyphrase Extraction from Disaster-related Tweets

While keyphrase extraction has received considerable attention in recent years, relatively few studies exist on extracting keyphrases from social media platforms such as Twitter, and even fewer for extracting disaster-related keyphrases from such sources. During a disaster, keyphrases can be extremely useful for filtering relevant tweets that can enhance situational awareness. Previously, joint training of two different layers of a stacked Recurrent Neural Network for keyword discovery and keyphrase extraction had been shown to be effective in extracting keyphrases from general Twitter data. We improve the model's performance on both general Twitter data and disaster-related Twitter data by incorporating contextual word embeddings, POS-tags, phonetics, and phonological features. Moreover, we discuss the shortcomings of the often used F1-measure for evaluating the quality of predicted keyphrases with respect to the ground truth annotations. Instead of the F1-measure, we propose the use of embedding-based metrics to better capture the correctness of the predicted keyphrases. In addition, we also present a novel extension of an embedding-based metric. The extension allows one to better control the penalty for the difference in the number of ground-truth and predicted keyphrases

cs.IR↗

arXiv2018

Localizing and Quantifying Damage in Social Media Images

Traditional post-disaster assessment of damage heavily relies on expensive GIS data, especially remote sensing image data. In recent years, social media has become a rich source of disaster information that may be useful in assessing damage at a lower cost. Such information includes text (e.g., tweets) or images posted by eyewitnesses of a disaster. Most of the existing research explores the use of text in identifying situational awareness information useful for disaster response teams. The use of social media images to assess disaster damage is limited. In this paper, we propose a novel approach, based on convolutional neural networks and class activation maps, to locate damage in a disaster image and to quantify the degree of the damage. Our proposed approach enables the use of social network images for post-disaster damage assessment and provides an inexpensive and feasible alternative to the more expensive GIS approach.

cs.CV↗