Daniel Guzman Olivares, Lara Quijano, Federico Liberatore. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Abstract This paper addresses the growing integration of machine learning and computer vision with a particular emphasis on classifying facial features. Although these technologies hold great potential, there are concerns regarding biases that must be mitigated to ensure more impartial results. This study primarily addresses the issue of bias caused by class imbalance, which can be mitigated through algorithmic or data-level approaches. Notably, the literature presents gaps, including the lack of comprehensive studies comparing these two types of mitigation techniques and understanding the circumstances in which each is more suitable. Moreover, the influence of imbalance conditions and data complexity on mitigation methods remains underexplored. To address these gaps, this research formulates three key research questions and conducts experiments using two datasets, UTKFace and PlantVillage, known for varying complexities. Various imbalance scenarios are simulated in these datasets. Additionally, a novel algorithm-level mitigation method named “Diffuse Focal Loss” is introduced. Results indicate the high effectiveness of synthetic oversampling methods, specifically using the Wasserstein variant of Generative Adversarial Networks, compared to algorithm-level approaches. Among the latter, the proposed novel method outperforms others in terms of metrics. However, it is worth noting that algorithmic techniques are more practical and quicker to apply in low-complexity scenarios.

Active investing strategies have poor historical long-term performance compared to passive strategies. Furthermore, many active strategies use forecasting of various signals even though the efficient market hypothesis posits that stock prices rapidly incorporate information. Despite this, recent studies continue to explore the use of forecast-based strategies and report positive evaluation results. We closely investigate several inter-day active management strategies based on forecasting models via a test set and a U.S. market backtest. We approximated transaction costs via bid and ask prices and percentage costs. Test metrics indicated that models could not forecast over week-long horizons but demonstrated some forecasting ability over monthly and quarterly horizons. Backtest results indicate a null correlation between test metrics and strategy performance. For low transaction cost rates, several strategies achieved higher Sharpe ratios than a benchmark passive strategy. At 25 basis points a single strategy had comparable performance to the benchmark. It is unclear to what extent this observation of superior strategy performance was due to the quantity of strategies we evaluated. Our findings emphasise the importance of evaluating strategies under realistic trading conditions and suggest further investigation into the null correlation between forecast accuracy and strategy performance.

Social media platforms have evolved into an online representation of our social interactions. We may use the resources they provide to analyze phenomena that occur within them, such as the development and viralization of offensive and hostile content. In today’s polarized world, the escalating nature of this behavior is cause for concern in modern society. This research includes an in-depth examination of previous efforts and strategies for detecting and preventing hateful content on the social network Twitter, as well as a novel classification approach based on users’ profiles, related social environment and generated tweets. This paper’s contribution is threefold: (i) an improvement in the performance of the HaterNet algorithm, an expert system developed in collaboration with the Spanish National Office Against Hate Crimes of the Spanish State Secretariat for Security (Ministry of the Interior) that is capable of identifying and monitoring the evolution of hate speech on Twitter using an LTSM + MLP neural network architecture. To that end, a model based on BERT, HaterBERT, has been created and tested using HaterNet’s public dataset, providing results that show a significant improvement; (ii) A methodology to create a user database in the form of a relational network to infer textual and centrality features. This contribution, SocialGraph, has been independently tested with various traditional Machine Learning and Deep Learning algorithms, demonstrating its usefulness in spotting haters; (iii) a final model, SocialHaterBERT, that integrates the previous two approaches by analyzing features other than those inherent in the text. Experiment results reveal that this last contribution greatly improves outcomes, establishing a new field of study that transcends textual boundaries, paving the way for future research in coupled models from a diachronic and dynamic perspective.

Term weighting schemes are widely used in Natural Language Processing and Information Retrieval.In particular, term weighting is the basis for keyword extraction.However, there are relatively few evaluation studies that shed light about the strengths and shortcomings of each weighting scheme.In fact, in most cases researchers and practitioners resort to the wellknown tf-idf as default, despite the existence of other suitable alternatives, including graphbased models.In this paper, we perform an exhaustive and large-scale empirical comparison of both statistical and graph-based term weighting methods in the context of keyword extraction.Our analysis reveals some interesting findings such as the advantages of the lessknown lexical specificity with respect to tf-idf, or the qualitative differences between statistical and graph-based methods.Finally, based on our findings we discuss and devise some suggestions for practitioners. 1

In the last year, thanks to the Ms. Pac-Man Versus Ghosts Competition, the game of Ms. Pac-Man has gained increasing attention from academics in the field of computational intelligence. In this paper, we contribute to this research stream by presenting a simple genetic algorithm with lexicographic ranking (GALR) for the optimization of flocking strategy-based ghost controllers. Flocking strategies are a paradigm for intelligent agents characterized by showing emergent behavior and for having very little computational and memory requirements, making them well suited for commercial applications and mobile devices. In particular, we study empirically the effect of optimizing homogeneous and heterogeneous teams. The computational analysis shows that the flocking strategy-based controllers generated by the proposed GALR outperform the ghost controllers included in the competition framework and some of those presented in the literature.