Gema Alcaráz-Mármol


2023

pdf bib
UMUTeam at SemEval-2023 Task 12: Ensemble Learning of LLMs applied to Sentiment Analysis for Low-resource African Languages
José Antonio García-Díaz | Camilo Caparros-laiz | Ángela Almela | Gema Alcaráz-Mármol | María José Marín-Pérez | Rafael Valencia-García
Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023)

These working notes summarize the participation of the UMUTeam in the SemEval 2023 shared task: AfriSenti, focused on Sentiment Analysis in several African languages. Two subtasks are proposed, one in which each language is considered separately and another one in which all languages are merged. Our proposal to solve both subtasks is grounded on the combination of features extracted from several multilingual Large Language Models and a subset of language-independent linguistic features. Our best results are achieved with the African languages less represented in the training set: Xitsonga, a Mozambique dialect, with a weighted f1-score of 54.89\%; Algerian Arabic, with a weighted f1-score of 68.52\%; Swahili, with a weighted f1-score of 60.52\%; and Twi, with a weighted f1-score of 71.14%.

pdf bib
UMUTeam at SemEval-2023 Task 11: Ensemble Learning applied to Binary Supervised Classifiers with disagreements
José Antonio García-Díaz | Ronghao Pan | Gema Alcaráz-Mármol | María José Marín-Pérez | Rafael Valencia-García
Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023)

This paper describes the participation of the UMUTeam in the Learning With Disagreements (Le-Wi-Di) shared task proposed at SemEval 2023, which objective is the development of supervised automatic classifiers that consider, during training, the agreements and disagreements among the annotators of the datasets. Specifically, this edition includes a multilingual dataset. Our proposal is grounded on the development of ensemble learning classifiers that combine the outputs of several Large Language Models. Our proposal ranked position 18 of a total of 30 participants. However, our proposal did not incorporate the information about the disagreements. In contrast, we compare the performance of building several classifiers for each dataset separately with a merged dataset.