Pasar al contenido principal

PLN CENTRADO EN EL SER HUMANO E IMPULSADA POR EL CONOCIMIENTO PARA GARANTIZAR LA CALIDAD, INTEGRIDAD Y TRANSPARENCIA EN UN ECOSISTEMA DIGITAL DE CONFIANZA

PID2024-156263OB-C22

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal para la Investigación y el Desarrollo Experimental
Subprograma Subprograma Estatal de Generación de Conocimiento Científico-Técnico y Desarrollo Experimental
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2024
Plan Estatal 2024-2027 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario UNIVERSIDAD DE ALICANTE

Resultados relacionados

Resultados totales (Incluyendo duplicados): 8
Encontrada(s) 1 página(s)

RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Span-Reference and Grounding Reliability in Generative Spanish Clinical Named Entity Recognition: A Validation Study

  • Grande Ruiz, Eduardo
  • Muñoz, Rafael
  • Gutiérrez, Yoan
  • Saquete Boró, Estela
  • Grande Ruiz, Eduardo|||/items/d0bc54e8-b11a-41c0-9f6b-3889b4ed8179
  • Muñoz, Rafael|||/items/d6ef1286-fc50-4e8b-9d97-bcbc30e3d1bc
  • Gutiérrez, Yoan|||/items/8f7f89db-a015-4ed6-85b9-75bd041ee8bd
  • Saquete Boró, Estela|||/items/452393f2-fb8d-4840-b797-4b65a4bc9a87
Generative named entity recognition (NER) systems must identify clinical concepts and map them to exact source spans. We investigated how span-reference design affects this mapping and whether failures arise from recognition or source localisation. Four representation-and-grounding pipelines—inline XML, mention-list JSON, a tab-separated mention list, and direct-offset JSON—were compared for disease, procedure, and symptom recognition in a public Spanish clinical case-report collection. Two instruction-tuned 3-billion-parameter models were fine-tuned for each task, and outputs were evaluated for syntactic validity, parsing, grounding, and official strict-span performance. Inline XML, mention-list JSON, and the tab-separated format achieved F1 scores of 0.637–0.736, whereas direct-offset JSON achieved 0.001–0.005. With the same unadapted checkpoints, zero-shot F1 was at most 0.131 and three-shot F1 at most 0.386; three-shot prompting improved the mention-list outputs, while inline XML and direct offsets remained near zero. Direct-offset outputs usually contained relevant mention text but incorrect character positions. On a separate 75-document confirmation partition, grounding the emitted strings recovered F1 of 0.603–0.701, while replacing absolute offsets with mention-occurrence numbers achieved 0.652–0.740. Without further training, the external CARMEN-I evaluation on 458 hospital-record sections yielded F1 of 0.610–0.659 for string-grounded or occurrence-index outputs, compared with 0.113–0.177 for inline XML and at most 0.002 for direct offsets. These results show that output validity alone does not establish usable span extraction. Separating recognition from source localisation identifies where an otherwise valid generation fails, while occurrence-based references provide a more reliable alternative to absolute character offsets under exact matching in the tested settings.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Nuanced Toxicity Detection in Spanish: A New Corpus and Benchmark Study

  • Mármol-Romero, Alba María
  • Sepúlveda-Torres, Robiert
  • Saquete Boró, Estela
  • Martín Valdivia, María Teresa
  • Ureña López, Luis Alfonso
  • Sepúlveda-Torres, Robiert|||/items/60f24ee2-fe45-4b5c-a70c-c025ed1b5e4a
  • Saquete Boró, Estela|||/items/452393f2-fb8d-4840-b797-4b65a4bc9a87
The rise of toxic content on digital platforms has intensified the demand for automatic moderation tools. While English has benefited from large-scale annotated corpora, Spanish remains under-resourced, particularly for nuanced cases of toxicity such as irony, sarcasm, or indirect aggression. We present an extended version of the NECOS-TOX corpus, comprising 4,011 Spanish comments collected from 16 major news outlets. Each comment is annotated across three levels of toxicity (Non-Toxic, Slightly Toxic, and Toxic), following an iterative annotation protocol that achieved substantial inter-annotator agreement (k = 0.74). To reduce annotation costs while maintaining quality, we employed a human-in-the-loop active learning strategy, with manual correction of model pre-labels. We benchmarked the dataset with traditional machine learning (ML) methods, domain-specific transformers, and instruction-tuned large language models (LLMs). Results show that compact encoder models (e.g., RoBERTa-base-bne, 125M parameters) perform on par with much larger models (e.g., LLaMA-3.1-8B), underscoring the value of in-domain adaptation over raw scale. Our error analysis highlights persistent challenges in distinguishing subtle forms of toxicity, especially sarcasm and implicit insults, and reveals entity-related biases that motivate anonymization strategies. The dataset and trained models are released publicly.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Exploring Cross-Lingual Transfer in Transformer-Based Fraud Detection Models

  • Martínez-Murillo, Iván
  • Sepúlveda-Torres, Robiert
  • Consuegra-Ayala, Juan Pablo
  • Martínez-Murillo, Iván|||/items/10185dd6-7baf-4c77-9241-9ca9c38f31a1
  • Sepúlveda-Torres, Robiert|||/items/60f24ee2-fe45-4b5c-a70c-c025ed1b5e4a
  • Consuegra-Ayala, Juan Pablo|||/items/5a98e4ca-607f-4aa0-91c2-a5d929bab52c
Digital communication channels have become major vectors for large-scale cyber fraud, including spam, phishing, and recruitment scams, causing significant financial losses and eroding user trust. While transformer-based models have improved text classification, most systems remain English-centric, limiting their effectiveness in multilingual environments and across diverse fraud typologies. In this work, we hypothesize that fraud detection models trained in a single language can exhibit cross-lingual generalization capabilities. We investigate the use of Natural Language Processing (NLP) for detecting multiple forms of textual fraud by developing and evaluating three transformer-based discriminative models targeting spam, phishing, and fake job postings. Specifically, we compare two architectures—MrBERT (308M parameters) and mROBERTa (283M parameters)—across English, Spanish, and Valencian data. Our results demonstrate that while models trained exclusively on English achieve near-perfect performance in-language (F1 ≈ 0.98), they exhibit limited cross-lingual generalization in other languages such as Spanish (F1 ≈ 0.44 for phishing detection, and F1 ≈ 0.85 for spam detection). However, incorporating multilingual training data dramatically improves performance in target languages (F1 ≈ 0.95–0.99) while maintaining or even enhancing English accuracy. We further observe that cross-lingual transfer is most effective when datasets are parallel or closely aligned. These findings underscore the critical importance of multilingual data inclusion for building robust, scalable cybersecurity frameworks in diverse linguistic settings.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Discriminative Benchmarking of Spanish Language Models: Findings from the ODESIA Challenge 2024, Evaluación Discriminativa de Modelos de Lenguaje en Español: Resultados del ODESIA Challenge 2024

  • Benito-Santos, Alejandro
  • Morante Vallejo, Roser
  • Ghajari, Adrián
  • García-Ferrero, Iker
  • Sepúlveda-Torres, Robiert
  • Rigau Claramunt, German
  • Agerri Gascón, Rodrigo
  • Consuegra-Ayala, Juan Pablo
  • Estevanell-Valladares, Ernesto L.
  • Yáñez Romero, Fabio
  • Canal-Esteve, Miquel
  • Gutiérrez, Yoan
  • Muñoz, Rafael
  • Palomar, Manuel
  • Sánchez Salido, Eva
  • Marco, Guillermo
  • Fernández García, Andrés
  • Fresno Fernández, Víctor
  • Amigó Cabrera, Enrique
  • Plaza Morales, Laura
  • Carrillo-de-Albornoz, Jorge
  • Lucas, Miguel
  • Gonzalo Arroyo, Julio
  • Gutiérrez, Yoan|||/items/8f7f89db-a015-4ed6-85b9-75bd041ee8bd
  • Muñoz, Rafael|||/items/d6ef1286-fc50-4e8b-9d97-bcbc30e3d1bc
  • Palomar, Manuel|||/items/71e8f2d8-8d56-4c8e-89df-d6ce2497d967
  • Consuegra-Ayala, Juan Pablo|||/items/5a98e4ca-607f-4aa0-91c2-a5d929bab52c
  • Sepúlveda-Torres, Robiert|||/items/60f24ee2-fe45-4b5c-a70c-c025ed1b5e4a
  • Estevanell-Valladares, Ernesto L.|||/items/d8b1af9a-112f-48de-813f-b4d8f1df8bd1
  • Canal-Esteve, Miquel|||/items/f1b086bf-bce4-49ea-b765-73081f7a27f9
  • Yáñez Romero, Fabio|||/items/792a2bc4-9249-45bf-8389-7a63e1b74671
This paper presents the results from the 2024 ODESIA Challenge, a public competition aimed at benchmarking natural language processing (NLP) systems in Spanish across ten discriminative tasks using a standardized methodology based on private, held-out test sets. Results show the winning system (Qwen2.5-14B) prevailed due to structural advantages in extractive Question Answering, whereas encoders outperformed LLMs in other tasks such as sequence labeling and soft classification. We conclude that, while generative models may dominate reasoning-heavy tasks involving long contexts, encoder architectures obtain on-par or even better performance in many other discriminative scenarios, challenging the assumption that massive scale universally supersedes specialized architectural design.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

EATITALL: An Artificial Intelligence Platform for Designing and Developing New Healthy Products in the Agri-Food Sector

  • Gutiérrez, Yoan
  • Muñoz, Rafael
  • Canal-Esteve, Miquel
  • Bonora, Mar
  • Cheikh-Moussa, Kamila
  • Gutiérrez, Yoan|||/items/8f7f89db-a015-4ed6-85b9-75bd041ee8bd
  • Muñoz, Rafael|||/items/d6ef1286-fc50-4e8b-9d97-bcbc30e3d1bc
  • Cheikh-Moussa, Kamila|||/items/48b844a7-e876-44b8-9235-d731557b07a6
  • Canal-Esteve, Miquel|||/items/f1b086bf-bce4-49ea-b765-73081f7a27f9
  • Bonora, Mar|||/items/9fffa829-b6b6-42d9-a4f0-671a7e25cbf8
This project proposes industrial research leveraging Artificial Intelligence technologies, particularly in the field of Natural Language Processing, to optimise the development of ingredients, formulations, and functional foods. It will do so through a platform designed to identify consumers’ nutritional needs by learning from clinical Gold Standard data. The EATITALL tool will apply this clinically validated knowledge to help agri-food companies better understand the nutritional requirements of the general population. The insights gained will support the development and marketing of product lines aimed at promoting healthier diets and preventing cardiometabolic diseases. The monitoring of physiological parameters and personalised follow-up during the clinical trial phase will enable the validation of newly developed formulations. Based on the results, these products may enter a cycle of continuous improvement, serving as a reference model for other companies in the agri-food sector.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Overview of GenSIE at IberLEF 2026: Schema-Guided Information Extraction with Small Language Models in Spanish, Resumen de GenSIE en IberLEF 2026: extracción de información guiada por esquemas con modelos de lenguaje pequeños en español

  • Piad-Morffis, Alejandro
  • Almeida-Cruz, Yudivian
  • Estévez-Velarde, Suilan
  • Espinosa-Zaragoza, Isabel|||/items/b9b30d7e-4b6b-4a40-b4c6-214bb9ba331f
  • Miró Maestre, María|||/items/54fede9e-4116-446d-b668-9871db0ac7f3
  • Sevilla Requena, Lucía|||/items/2ef33fba-eec8-4cea-a86c-171f04605aaf
  • Pérez Montero, Alba|||/items/933a3a2a-82b5-4d6e-9337-2e478a6c3f42
  • Estevanell-Valladares, Ernesto L.|||/items/d8b1af9a-112f-48de-813f-b4d8f1df8bd1
GenSIE is the IberLEF 2026 shared task on schema-guided information extraction from Spanish text with small language models. A CPU-only system receives a (text, instruction, JSON schema) triple and must emit a schema-conformant object faithful to the text, returning null for fields the text does not support. We contribute a 271-instance Spanish development benchmark over eight domains, a 145-instance held-out gold test set that adds a ninth, a semi-automatic agentic curation pipeline with three human gates, and an evaluator pairing micro-F1 over flattened schemas with a multi-model gap-closed metric. Eleven teams submitted thirty-five pipelines, scored on 125 instances against no-think baselines (Gemma 4 E4B F1=0.7780, Qwen3-14B F1=0.7542). DRILLER leads, closing 27.7% of the baseline-to-perfect gap, though a paired bootstrap cannot separate the top three. Retrieval and schema enrichment pay off; extra inference passes do not.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

XAutoLM: Efficient Fine-Tuning of Language Models via Meta-Learning and AutoML

  • Estevanell-Valladares, Ernesto L.
  • Estévez-Velarde, Suilan
  • Gutiérrez, Yoan
  • Montoyo, Andres
  • Mitkov, Ruslan
  • Estevanell-Valladares, Ernesto L.|||/items/d8b1af9a-112f-48de-813f-b4d8f1df8bd1
  • Gutiérrez, Yoan|||/items/8f7f89db-a015-4ed6-85b9-75bd041ee8bd
  • Montoyo, Andres|||/items/a2b7158a-7032-423a-a7b5-0f7b92902b72
  • Mitkov, Ruslan|||/items/30b15727-6f78-421c-ba49-df42c02f0e96
Experts in machine learning leverage domain knowledge to navigate decisions in model selection, hyperparameter optimization, and resource allocation. This is particularly critical for fine-tuning language models (LMs), where repeated trials incur substantial computational overhead and environmental impact. However, no existing automated framework simultaneously tackles the entire model selection and hyperparameter optimization (HPO) task for resource-efficient LM fine-tuning. We introduce XAutoLM, a meta-learning-augmented AutoML framework that reuses past experiences to optimize discriminative and generative LM fine-tuning pipelines efficiently. XAutoLM learns from stored successes and failures by extracting task- and system-level meta-features to bias its sampling toward valuable configurations and away from costly dead ends. On four text classification and two question-answering benchmarks, XAutoLM surpasses zero-shot optimizer’s peak F1 on five of six tasks, cuts mean evaluation time of pipelines by up to 4.5x, reduces search error ratios by up to sevenfold, and uncovers up to 50% more pipelines above the zero-shot Pareto front. In contrast, simpler memory-based baselines suffer negative transfer. We release XAutoLM and our experience store to catalyze resource-efficient, Green AI fine-tuning in the NLP community.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Quality over Quantity: Ontology Dataset Selection for Efficient Triple Generation in LLMs, Calidad frente a cantidad: selecci´on de datasets ontol´ogicos para la generaci´on eficiente de triples en LLMs

  • Canal-Esteve, Miquel
  • Canal-Esteve, Miquel|||/items/f1b086bf-bce4-49ea-b765-73081f7a27f9
  • Abreu Salas, José Ignacio|||/items/99fd9ec6-0b97-4c02-a28e-98bc309cff48
  • Gutiérrez, Yoan|||/items/8f7f89db-a015-4ed6-85b9-75bd041ee8bd
  • Muñoz, Rafael|||/items/d6ef1286-fc50-4e8b-9d97-bcbc30e3d1bc
The integration of structured semantic knowledge into Large Language Models (LLMs) is increasingly relevant for knowledge-intensive Natural Language Processing tasks. Ontologies play a central role in organizing this knowledge, yet their manual construction and maintenance become inefficient as digital information grows. While LLMs are increasingly used for ontology learning, their ability to generate coherent structural knowledge remains limited, particularly in small open-source models needed for efficient and sustainable deployment. This work investigates a controlled ontology-to-ontology generation setup to evaluate LLMs’ structural generalization abilities and introduces a data-centric continual pretraining methodology based on ontology quality. Two open-source models, LLaMA 3.2-1B and GEMMA 3-1B, are trained and evaluated using structural and formatting criteria. Results show that selecting only high-quality ontology data substantially improves triple generation while reducing training time by an average of 67% for 1 epoch.