https://reunir.unir.net/handle/123456789/19139 2026-03-20T18:29:41Z https://reunir.unir.net/handle/123456789/19169 Predicting Tree Growth and Transpiration in Forests: An Analysis of a Small-Scale Dataset With Pareto Optimized Tsaug Augmentation Maskeliūnas, Rytis; Damaševičius, Robertas; Odusam, Modupe; Sidabrienė, Diana; Augustaitis, Algirdas; Mozgeris, Gintautas The study demonstrates the potential of specifically developed data augmentation in estimating tree growth and transpiration by emphasizing the influence of environmental variables, such as photosynthetically active radiation (PAR), air temperature, and relative humidity—on tree growth predictions. The investigation utilizes data obtained from two hemi-boreal semi-natural mixed conifer deciduous forest sites in the Aukstaitija National Park in Lithuania. Field measurements included xylem sap flow measurements and stem circumference increment growth. The dataset utilized in the analysis consisted of four trees per species and contained information on tree growth, transpiration, and solar angle measurements. Pareto-optimized Tsaug augmentation techniques were employed to diversify the dataset, generating augmented time series to improve diversity and minimize distortion. The results of the correlation analysis indicated significant relationships between environmental variables and tree growth and transpiration. The Prophet based prediction model, notably when trained with augmented data, outperformed other models in predicting tree growth and perspiration variables (MAPE ranging from 0.0017 to 0.01). This was particularly evident for FACP, FAGP, and FADP variables, showcasing substantial improvement with augmented data. Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-10T08:10:32Z No. of bitstreams: 2 Predicting Tree Growth and Transpiration in Forests.pdf: 6350547 bytes, checksum: ffc4a4d5d11641ba99b287683fe72a1d (MD5) Predicting Tree Growth and Transpiration in Forests.pdf: 6350547 bytes, checksum: ffc4a4d5d11641ba99b287683fe72a1d (MD5); Made available in DSpace on 2026-03-10T08:10:32Z (GMT). No. of bitstreams: 2 Predicting Tree Growth and Transpiration in Forests.pdf: 6350547 bytes, checksum: ffc4a4d5d11641ba99b287683fe72a1d (MD5) Predicting Tree Growth and Transpiration in Forests.pdf: 6350547 bytes, checksum: ffc4a4d5d11641ba99b287683fe72a1d (MD5) https://reunir.unir.net/handle/123456789/19151 AI Powered Commentary and Camera Direction in E-Sports Narayanan, Swathi Jamjal; Joseph, Kevin Winston; Sirohi, Devansh; Chaudhary, Harsh; Shivkumar, Hitesh Real-time, AI-driven commentary and camera direction provide revolutionary possibilities to improve spectator engagement and comprehension of live events in the rapidly advancing world of e-sports. This paper proposes an autonomous system designed to both generate dynamic commentary as well as control the spectator camera for live-streamed e-sports matches, specifically focusing on League of Legends (LoL), a popular Multiplayer Online Battle Arena (MOBA) game. It incorporates the use of GPT-4o with Vision and OpenAI’s TTS API. Synchronization of commentary with real-time camera movements is one of the major challenges tackled. This is done using a camera tracking and scene change detection algorithm that effectively adjusts the commentary to changing scenes in real-time by utilizing computer vision techniques. Further, two neural architectures for AI-driven camera control: a 2D Convolutional-LSTM (Conv-LSTM) model that concentrates on independent spatial and temporal analysis, and a 3D CNN model that combines these features to forecast camera movements in a more comprehensive way are presented. Evaluations on fluency, relevance, and strategic depth metrics, show that our integrated system improves viewer experience by providing deep and coherent narratives that are contextually aligned with the game dynamics. The proposed models are evaluated quantitatively in capturing spectator camera movement patterns. Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-09T16:21:30Z No. of bitstreams: 1 AI Powered Commentary and Camera Direction in E-Sports.pdf: 3709678 bytes, checksum: aed05a810871b511f14e69d84e5d5236 (MD5); Made available in DSpace on 2026-03-09T16:21:30Z (GMT). No. of bitstreams: 1 AI Powered Commentary and Camera Direction in E-Sports.pdf: 3709678 bytes, checksum: aed05a810871b511f14e69d84e5d5236 (MD5) https://reunir.unir.net/handle/123456789/19150 The Application of Large Language Models and Virtual Assistants in Project Management Research: A Review Gil Ruiz, Jesús; Zayas-Gallardo, Javier; Díaz Rodríguez, Hernán The rapid evolution of generative artificial intelligence (AI) is transforming project management practices by enhancing efficiency, productivity and adaptability in decision-making processes. The integration of large language models (LLMs) into project management research and practice is reviewed, with a particular focus on virtual assistants as decision support tools. State-of-the-art models such as Mistral, Large Language Model Meta AI (LLaMa), Bidirectional Encoder Representations from Transformers (BERT) and T5, are assessed for their potential to automate complex project tasks, extract insights from project datasets, and optimize decision-making across various project management domains and business sectors. Generative AI is shown to surpass traditional project management systems by not only analysing historical project data but also generating new strategies and solutions in real time. Applications include project risk assessment, resource allocation optimisation, stakeholder communication and project performance prediction. The role of fine-tuning and retraining LLMs to adapt them to industry-specific project management challenges is also examined enhancing relevance and performance across diverse business environments. Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-09T15:06:11Z No. of bitstreams: 1 The Application of Large Language Models and Virtual Assistants in Project Management Research.pdf: 652904 bytes, checksum: 50aafbb737b9685e39f76a1e23b5e7dd (MD5); Made available in DSpace on 2026-03-09T15:06:11Z (GMT). No. of bitstreams: 1 The Application of Large Language Models and Virtual Assistants in Project Management Research.pdf: 652904 bytes, checksum: 50aafbb737b9685e39f76a1e23b5e7dd (MD5) https://reunir.unir.net/handle/123456789/19149 Blending Language Models and Domain-Specific Languages in Computer Science Education. A Case Study on API RESTFul Jurado, Francisco; Rodríguez, Francy D.; Chavarriaga, Enrique; Rojas, Luis Since Computer Science students are used to applying both General Purpose Programming Languages (GPPLs) and Domain-Specific Languages (DSLs), Generative Artificial Intelligence based on Language Models (LMs) can help them on automatic tasks, allowing them to focus on more creative tasks and higher skills. However, the teaching and evaluation of technical tasks in Computer Science can be inefficient and prone to errors. Thus, the main objective of this article is to explore the performance of LMs compared to that of undergraduate Computer Science students in a specific case study: designing and implementing RESTful APIs DSLs. This research aims to determine if LMs can enhance the efficiency and accuracy of these processes. Our case study involved 39 students and 5 different LMs that must use the two DSLs we also designed for their task assignment. To evaluate performance, we applied uniform criteria to student and LMs-generated solutions, enabling a comparative analysis of accuracy and effectiveness. With a case study comparing performance between students and LMs, this article contributes to checking to what extent LMs are able to carry out software development tasks involving the use of new DSLs specially designed for highly specific settings in a similar way as well-qualified Computer Science students are able to. The results underscore the importance of welldefined DSLs and effective prompting processes for optimal LM performance. Specifically, LMs demonstrated high variability in task execution, with two GPT-based LMs achieving similar grades to those scored by the best of the students for every task, obtaining 0.78 and 0.92 on a normalized scale [0, 1], with 0.23 and 0.14 Standard Deviation for ChatGPT-4 and ChatGPT-4o respectively. After the experience, we can conclude that a well-defined DSL and a proper prompting process, providing the LM with metadata, persistent prompts, and a good knowledge base, are crucial for good LM performance. When LMs receive the right prompts, both large and small LMs can achieve excellent results depending on the task. Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-09T14:48:50Z No. of bitstreams: 1 Blending Language Models and Domain-Specific Languages in Computer Science Education. A Case Study on API RESTFul.pdf: 4031896 bytes, checksum: 5fc863a540c8e5b936bbe5966bcf8669 (MD5); Made available in DSpace on 2026-03-09T14:48:50Z (GMT). No. of bitstreams: 1 Blending Language Models and Domain-Specific Languages in Computer Science Education. A Case Study on API RESTFul.pdf: 4031896 bytes, checksum: 5fc863a540c8e5b936bbe5966bcf8669 (MD5) https://reunir.unir.net/handle/123456789/19148 AI Prediction and Teaching Strategies for a Two-Phase Engine in a Smart Learning Platform Real-Fernández, Alberto; García-Sigüenza, Javier; Llorens-Largo, Faraón; Molina-Carmona, Rafael The impact and progress of Information Technologies has led to a process of change in most environments of our society, specially education. Even more with the current rise of Artificial Intelligence, what has led to the creation of different new tools aiming to improve the learning experience. This fact has contributed to the creation of systems that aim to adapt the learning process to each individual learner and offer them a personalised experience. The problem of letting automated systems manage the whole learning process is the lack of human factor, but learning objectives and teacher criteria are crucial. That is why this research proposes a solution that combines the potential of AI without neglecting the teacher decision. Concretely, the proposal is an AI model that selects the most suitable activity to each learner. To do so, this proposed model is structured in two phases. The first is the prediction phase, in which the model predicts the score a learner will obtain and the time they will spend to complete an activity. Then, in the second phase, the selection of a single activity is done by means of instructional strategies. These strategies are based on the previously obtained metrics and establish the criteria to follow for selecting activities. The selected strategy is always set by the teacher, who will guide the learners through the process. With this model, this research proposes a combination of AI techniques with human decision-making. Instead of relying the learning process to an automated engine, it includes the teacher as the one to guide the AI by making the last decision. Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-09T14:45:05Z No. of bitstreams: 1 AI Prediction and Teaching Strategies for a Two-PhaseEngine in a Smart Learning Platform.pdf: 1755160 bytes, checksum: 2e86e87974bd89b61f9a1eb045ed5776 (MD5); Made available in DSpace on 2026-03-09T14:45:05Z (GMT). No. of bitstreams: 1 AI Prediction and Teaching Strategies for a Two-PhaseEngine in a Smart Learning Platform.pdf: 1755160 bytes, checksum: 2e86e87974bd89b61f9a1eb045ed5776 (MD5) https://reunir.unir.net/handle/123456789/19147 UPMVM: A Metrics Verification Model for Urdu Poetry Zaman, Asia; Ud-Din, Zia; Iqbal, Sajid; Al Shuhail, Asma Urdu poetry retains a prominent position in the cultural heritage of Urdu language. Rhyme schemes and meters are frequently employed in poetry, which follow specific patterns and structures. Natural Language Processing has the capacity to recognize and analyze these patterns, which is beneficial in the investigation of poetic forms. This research presents the UPMVM (Urdu Poetry Metrics Verification Model), a novel rulebased architecture, designed for detecting meter of any given Urdu ghazal verse. In this work, we propose an algorithm that consists of sixteen steps that identifies the Arud meter in the Urdu verses using a custom developed system. This application will not only assist professional poets but also enable students to examine poetry within the framework of prosody principles. The accurate analysis of the prosody of any poetry relies on the act of uttering words rather than on a written record. UPMVM consists of two phases: 1) The primary objective of the initial phase is to consolidate all available literature of the Arud system into a unified digital platform, then develop individual and combined DFA of each identified meter for pattern recognition; 2) the second phase is about the algorithmic implementation. All these rhythmical patterns are matched with 290 Arud meters and their sub-meters developed during this study. The implementation strategy of phase 2 comprises of five essential sub-phases including tokenization, orthography, syllable identification, weight assignment, and meter detection. For evaluation of the proposed method, three different datasets are used for feature extraction, token identification and performance measurement for identification of rhythmic patterns in Urdu poetry. The UPMVM model reached to promising outcome with an average accuracy of 94%. Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-09T14:36:58Z No. of bitstreams: 1 UPMVM A Metrics Verification Model for Urdu Poetry.pdf: 5797789 bytes, checksum: da7654d73fe50f6aaca78712a3fe24ac (MD5); Made available in DSpace on 2026-03-09T14:36:58Z (GMT). No. of bitstreams: 1 UPMVM A Metrics Verification Model for Urdu Poetry.pdf: 5797789 bytes, checksum: da7654d73fe50f6aaca78712a3fe24ac (MD5) https://reunir.unir.net/handle/123456789/19146 Multi-Class Dental CBCT Segmentation in Data- Constrained Scenarios Through Transformers Giménez-Aguilar, Rafael C.; Paraíso-Medina, Sergio; García-Remesal, Miguel; Pradíes Ramiro, Guillermo Jesús; Bonfanti-Gris, Monica; Alonso-Calvo, Raúl Accurate segmentation of dental structures from cone-beam computed tomography (CBCT) images has become an active research field due to the widespread use of this technology in clinical practice. In recent years, contributions have shifted from traditional computer vision methods to deep learning-based approaches. However, most of these works are based solely on convolutional neural networks (CNNs), whereas the image segmentation state-of-the-art is currently moving towards attention-based architectures. Furthermore, contributions on dental CBCTs predominantly present methods focused on a single object category, mainly teeth. In this article we tackle the segmentation of multiple oral structures by implementing previously unutilized query-based segmentation transformers. The proposed method achieves similar results to the stateof- the-art, especially on tooth segmentation, while employing a considerably smaller training dataset than prior contributions. Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-09T14:33:05Z No. of bitstreams: 1 Multi-Class Dental CBCT Segmentation in Data-Constrained Scenarios Through Transformers.pdf: 2239302 bytes, checksum: 9add49d933551e3e343789ba9467739a (MD5); Made available in DSpace on 2026-03-09T14:33:05Z (GMT). No. of bitstreams: 1 Multi-Class Dental CBCT Segmentation in Data-Constrained Scenarios Through Transformers.pdf: 2239302 bytes, checksum: 9add49d933551e3e343789ba9467739a (MD5) https://reunir.unir.net/handle/123456789/19145 Robust Federated Learning With Contrastive Learning and Meta-Learning Zhang, Huan; Chen, Yuxian; Li, Kuanching; Li, Yuhui; Zhou, Sisi; Liang, Wei; Poniszewska-Maranda, Aneta Federated learning is regarded as an effective approach to addressing data privacy issues in the era of artificial intelligence. Still, it faces the challenges of unbalanced data distribution and client vulnerability to attacks. Current research solves these challenges but ignores the situation where abnormal updates account for a large proportion, which may cause the aggregated model to contain excessive abnormal information to deviate from the normal update direction, thereby reducing model performance. Some are not suitable for non-Independent and Identically Distribution (non IID) situations, which may lead to the lack of information on small category data under non-IID and, thus, inaccurate prediction. In this work, we propose a robust federated learning architecture, called FedCM, which integrates contrastive learning and meta-learning to mitigate the impact of poisoned client data on global model updates. The approach improves features by leveraging extracted data characteristics combined with the previous round of local models through contrastive learning to improve accuracy. Additionally, a meta-learning method based on Gaussian noise model parameters is employed to fine-tune the local model using a global model, addressing the challenges posed by non-independent and identically distributed data, thereby enhancing the model’s robustness. Experimental validation is conducted on real datasets, including CIFAR10, CIFAR100, and SVHN. The experimental results show that FedCM achieves the highest average model accuracy across all proportions of attacked clients. In the case of a non-IID distribution with a parameter of 0.5 on CIFAR10, under attack client proportions of 0.2, 0.5, and 0.8, FedCM improves the average accuracy compared to the baseline methods by 8.2%, 7.9%, and 4.6%, respectively. Across different proportions of attacked clients, FedCM achieves at least 4.6%, 5.2%, and 0.45% improvements in average accuracy on the CIFAR10, CIFAR100, and SVHN datasets, respectively. FedCM converges faster in all training groups, especially showing a clear advantage on the SVHN dataset, where the number of training rounds required for convergence is reduced by approximately 34.78% compared to other methods. Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-09T13:30:05Z No. of bitstreams: 1 Robust Federated Learning With Contrastive Learning and Meta-Learning.pdf: 3831396 bytes, checksum: 90f38925e9ef075b0d00a682cd267905 (MD5); Made available in DSpace on 2026-03-09T13:30:05Z (GMT). No. of bitstreams: 1 Robust Federated Learning With Contrastive Learning and Meta-Learning.pdf: 3831396 bytes, checksum: 90f38925e9ef075b0d00a682cd267905 (MD5) https://reunir.unir.net/handle/123456789/19144 PRESTO: A Recommender of Musical Collaborations Based on Heterogeneous Graph Neural Networks Terroso-Saenz, Fernando; Soto, Jesús; Muñoz, Andrés; Roose, Philippe The music industry is now more complex and competitive than ever before. In recent years, the search for collaborations with other artists has become a common strategy for musicians to maintain their presence in the sector. Besides, existing music streaming services such as Spotify have exposed large data feeds that can be used to develop innovative services within the realm of music. In this context, the present work introduces PRESTO, a novel recommendation system to suggest musicians for new collaborations with other artists by means of an ensemble of Graph Neural Networks. The system is fed with a heterogeneous graph representing the time evolution and the stationary aspects of a musician’s career. Finally, the proposal has been evaluated with a dataset comprising more than 200,000 artists, with an average F1 score above 0.75. Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-09T13:23:42Z No. of bitstreams: 1 PRESTO A Recommender of Musical Collaborations.pdf: 6405036 bytes, checksum: 43a5d995695acffdd554a64843b0e3e2 (MD5); Made available in DSpace on 2026-03-09T13:23:42Z (GMT). No. of bitstreams: 1 PRESTO A Recommender of Musical Collaborations.pdf: 6405036 bytes, checksum: 43a5d995695acffdd554a64843b0e3e2 (MD5) https://reunir.unir.net/handle/123456789/19143 Recommender Systems: Learning Collaborative Filtering Similarity Measures Using Siamese Networks Bobadilla, Jesús; Gutierrez, Abraham Improving current similarity measures in the collaborative filtering Recommender Systems is relevant, since it contributes to different applications such as to get better big data representations of users and items, to implement dynamic browsers able to navigate through data, and to explain recommendation results. Currently, there are many statistically based similarity measures, some of them tailored to the extraordinarily sparse collaborative filtering scenario. Nevertheless, the hypothesis of the paper is that using neural networks, learnt similarity measures can be obtained that improve existing ones. To accomplish the task, the typical neural models cannot be used, and it is necessary to focus on the similarity learning area, in which the goal is to make the model learn, which is a similarity function able to measure how similar two objects are. Siamese networks adequately implement the similarity learning concept, and we have adapted them to collaborative filtering particularities. The results in different scenarios show significant improvements compared to the state-of-the-art. Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-09T13:13:20Z No. of bitstreams: 1 Recommender Systems.pdf: 2150587 bytes, checksum: 4f500a3c2bc899557bf7e273e92cb677 (MD5); Made available in DSpace on 2026-03-09T13:13:20Z (GMT). No. of bitstreams: 1 Recommender Systems.pdf: 2150587 bytes, checksum: 4f500a3c2bc899557bf7e273e92cb677 (MD5) https://reunir.unir.net/handle/123456789/19142 Source Credibility Assessment in the Realm of Information Disorder: A Literature Review Cosentino, Alessia; de Maio, Carmen; Furno, Domenico; Gallo, Mariacristina; Loia, Vincenzo The proliferation of information disorder in the digital age has sparked a growing concern regarding the credibility of sources disseminating information. This review examines the evolving landscape of source credibility within information disorder. The review synthesizes key findings and trends related to the factors influencing source credibility, including available tools, shared indicators, and existing methods experimented with in calculating source credibility. The analysis highlights that from a more commercial point of view, several tools are aimed at analyzing the content’s credibility and studying the sources’ credibility. However, from a methodological point of view, there is still something more to do. Indicators that can be used to carry out a source credibility assessment focus on the structure and design of the source, excluding others indicating how the page traffic could be. As for the techniques to be used to assess the credibility of a source, it emerged that more innovative techniques, such as deep-learning, are being developed alongside slightly more classical statistical methods. The review analyzes 23 papers from Conferences and 22 from Journals published in recent years. It also identifies avenues for future inquiry and the development of effective strategies to combat the challenges posed by misinformation in the digital era. Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-09T13:09:15Z No. of bitstreams: 1 Source Credibility Assessment in the Realm of Information Disorder.pdf: 1288035 bytes, checksum: 288f3c7999f4fb1dd7915adaa340ceff (MD5); Made available in DSpace on 2026-03-09T13:09:15Z (GMT). No. of bitstreams: 1 Source Credibility Assessment in the Realm of Information Disorder.pdf: 1288035 bytes, checksum: 288f3c7999f4fb1dd7915adaa340ceff (MD5) https://reunir.unir.net/handle/123456789/19141 Editor's Note Verdú, Elena The International Journal of Interactive Multimedia and Artificial Intelligence (IJIMAI) provides a forum for researchers and professionals to share recent advances in artificial intelligence (AI) and its wide range of applications. This issue brings together contributions that reflect the growing diversity of AI research, covering topics such as information credibility assessment in social media, recommender systems, federated learning and data privacy, medical image analysis, educational technologies, large language models and virtual assistants, intelligent multimedia systems, and environmental monitoring. Collectively, these works illustrate how current AI methods continue to expand across disciplines, addressing both theoretical challenges and real-world problems. Downloads Submitted by Angela María Porras Ruiz (angela.porras@unir.net) on 2026-03-09T13:02:32Z No. of bitstreams: 1 Editor’s Note.pdf: 93685 bytes, checksum: 045066c7184207c043bbacf864d7a80c (MD5); Made available in DSpace on 2026-03-09T13:02:32Z (GMT). No. of bitstreams: 1 Editor’s Note.pdf: 93685 bytes, checksum: 045066c7184207c043bbacf864d7a80c (MD5)