Sistema Regional de Información
en línea para Revistas Científicas de América Latina,
el Caribe, España y Portugal

Software projects often accumulate technical debt, which undermines code quality, maintainability, and long-term viability. This article analyses twelve categories of technical debt and examines the capabilities of existing mitigation tools. To achieve this, we conducted a systematic literature review. Based on the results, we categorise and describe each debt type as it occurs in practice, and we assess prominent tools (such as AnaConDebt, CAST, DebtFlag, Visminer TD, and TD-Tracker) with respect to the debt types they target, the programming languages they support, and their adherence to established software quality models or methodologies. Our analysis shows that most current solutions focus on code and architectural debt, whereas documentation and process debt remain largely unaddressed. Moreover, most tools do not conform to software quality models or standards, which none of them explicitly incorporate.

The increasing prevalence of mental health disorders like depression and anxiety often leads to suicide, with individuals frequently expressing such thoughts on social media. Utilizing machine learning techniques to analyze social media texts would work towards preventing these outcomes, even though predicting suicide risk remains a challenge. In this study machine learning classifiers were developed aiming to detect suicide indicators using the Kaggle Suicide and Depression Detection dataset (Komati, N. Suicide Watch). Four models—Multinomial Naive Bayes, Gradient Boost Machine (GBM), Random Forest and Support Vector Machines—were tested, yielding promising results: Among the four models presented here SVM with a 0.95 Precision and 0.94 F1 score showed the best results. Keywords: Suicide, Supervised Learning, Machine Learning, Naïve Bayes, Gradient Boosting Machine, Random Forest, Support Vector Machines.

We propose a hybrid approach for multilingual sentiment analysis that combines extractive and abstractive summarization to address the limitations of standalone methods. The model integrates TF-IDF-based extraction with a fine-tuned XLM-R abstractive module, enhanced through dynamic thresholding and cultural adaptation. Experiments across 10 languages demonstrate significant improvements over baselines, achieving an accuracy of 0.90 for English and 0.84 for low-resource languages. The approach also achieves 22% greater computational efficiency compared to traditional methods. Practical applications include real-time brand monitoring and cross-cultural discourse analysis. Future work will focus on optimizing performance for low-resource languages through 8-bit quantization.

The SAT problem is important in the theory of computational complexity. It has been deeply studied because solutions for fragments of SAT can be transformed into solutions for several CSPs, including problems in areas such as Artificial Intelligence and Operations Research. Although SAT is an NP-complete problem, it is known that SAT is fixed-parameter tractable if we take any hypertree width as a parameter. In this work, we present several hypergraphs and countable classes of hypergraphs. For these classes of hypergraphs, we analyze their generalized hypertree width to prove that all the CSPs modeled with those hypergraphs are tractable.

This study explores an approach involving the adaptation of data sampling techniques within the hidden feature space of deep neural networks. By modifying traditional prototype selection and cleaning methods, our methodology eliminates noisy samples and condenses the data into representative points, thereby enhancing class separation and improving generalisation. A nearest-neighbours search in the hidden space enables more refined sample selection. Comprehensive experiments on four multi-class imbalanced hyperspectral datasets (Indian Pines, Salinas, PaviaU, and Pavia) demonstrated that combining over-sampling in the spectral space with editing in the hidden feature space outperforms conventional sampling methods. The best results were achieved with configurations such as ROS-TL-H2 and ROS-ENN-H3, which consistently yielded g-mean values above 0.90, showcasing the effectiveness of hidden-space editing. This strategy effectively balances class distributions while preserving informative samples, thereby improving classification performance and model robustness. Despite the increased computational complexity, the benefits justify its adoption in challenging scenarios involving class imbalance. The findings suggest that this approach may be particularly valuable for remote sensing and other highly imbalanced data classification tasks.

This research aims to integrate a Conversational System in virtual educational environments, reflecting the three dimensions of educational knowledge: know – know, know – do and know – be. Its purpose is to improve the personalization and quality of online learning through educational coaching. Although online interaction is increasingly relevant, current systems do not holistically integrate these three areas of knowledge. This study focuses on modeling "knowing – being", using speech act theory, Socratic dialectic and personality profile to generate responses that encourage active and reflective listening. Through natural language processing, the intelligent agent can interact more effectively, acting as a virtual coach that provides empathetic and personalized feedback. The research seeks to optimize virtual learning environments and advance the simulation of realistic interactions with virtual entities.

Feature selection is the process of extracting the most relevant features from a dataset, helping to reduce its dimensionality by eliminating non-essential features. This leads to simpler, faster models and optimises training efficiency. This paper presents two memetic algorithms: one employs a mono-objective filter method as a fitness function, while the other adopts a multi-objective approach. The latter uses the number of attributes in the dataset as the first objective, and the sum of Pearson’s correlations for the selected attributes as the second. Additionally, we apply a novel approach to the use of correlation for attribute selection within the aforementioned memetic algorithms. Both proposals aim to identify the most relevant attributes to reduce the dimensionality of twelve test datasets. The performance of the selected features was evaluated using a J48 decision tree. The results showed a reduction in the number of attributes ranging from 14% down to 5%, while accuracy varied from −5% up to 11%, with an average improvement of over 4% (considering only those datasets where accuracy changed).

Recent studies have explored the use of deep learning algorithms for the early detection of diseases. This work presents a comparative analysis of three state-of-the-art deep learning model architectures applied to this task. A study was conducted to test and compare these models using a dataset of 7,000 sugarcane leaf images categorized into five classes (healthy leaves and four disease types) and evaluated the performance of these models across various classification metrics to determine the most effective approach. Sugarcane is one of Mexico’s principal crops (INEGI, 2025), playing a crucial role in the sugar industry and its derivatives. However, various diseases pose a threat to sugarcane cultivation, resulting in significant economic losses due to the large-scale eradication of crops. Early and accurate identification of diseases is essential for effective management, yet it remains challenging without specialised knowledge. Deep learning tools can facilitate the detection of such diseases. This study presents a comparative analysis of three state-of-the-art deep learning architectures—EfficientNetV2B0, DenseNet121, and ResNet101V2—for sugarcane disease detection. Using a dataset of 7,000 sugarcane leaf images categorised into five classes (healthy and four disease types), the evaluation of these models was based on multiple classification metrics. The findings highlight competitive performance among the models, showcasing their respective strengths and limitations in terms of accuracy and computational efficiency. This analysis offers valuable insights into deep learning-based approaches for sugarcane disease detection, supporting the development of practical solutions for the agricultural sector.

Computer security policies are relatively new to many organisations, particularly regarding their Information Security Management Systems (ISMS). Although their conceptual origins date back to the 1980s, verifying these policies computationally remains a significant challenge. This research proposes a new tool for the verification of ISMS policies based on the VPD methodology. This methodology assesses information security policies and their compliance with ISMS by comparing the set of directive policies (M1) with the implemented policies (M2). The case study presented in this paper involves the M2 policies implemented in the security system of the Mayor’s Office in the municipality of Funza, Cundinamarca. These are based on established standards, such as ISO 27001, ITIL best practice libraries, the OISM3 guidelines, and Colombian government regulations—particularly those aligned with standards set by the Ministry of Information and Communication Technologies (MINTIC). The main contribution of this research is the development of ESVIT, an expert system built upon the VPD methodology to support the evaluation of policies in both public and private sector entities. 

This research addresses the problem of scheduling electric vehicle charging times, with the primary objective of minimising total tardiness, defined as the waiting time beyond the specified charging duration. The complexity arises from multiple interacting constraints, making it difficult to produce a feasible schedule that also minimises tardiness. As this problem is NP-hard, this study proposes a metaheuristic approach integrating a cellular processing algorithm with a Greedy Randomised Adaptive Search Procedure (GRASP). This paper provides a detailed implementation and description of the methods, along with a comprehensive calculation of the objective function, addressing areas that require further exploration in the existing literature.