David Griol’s research while affiliated with University of Granada and other places

... As in previous work [17,23], we have used the sklearn library to train several classifiers on the problem of negative thought detection using embeddings extracted from LMs. Specifically, we have trained the following models: SVM with 'RBF' or 'linear' kernel and C between 0.001 and 100 in steps multiples of ten (0.001, 0.01, 0.1, 1.0, 10, 100); logistic regression models with C between 0. 001 and 10 in steps multiples of 10; logistic regression models with C between 0. 001 and 10 in steps multiples of 10. 001 to 10 in steps multiples of 10; k-NN with neighbors ranging from 5 to 30 in steps of 5 (5,10,15,20,25,30); nearest center (NC) algorithm with default parameters; decision trees (DT) with the parameter min's parameter samples ranging from 5 to 30 in steps of 5; random forests (RF) with different estimators ranging from 10 to 90 in steps of 10; and MLP with 1, 2, or 3 layers of 80 neurons, respectively. ...

Reference:

Applying Transfer-Learning on Embeddings of Language Models for Negative Thoughts Classification

... Hence, smart contract validation offers a powerful tool for secure and automated transactions within blockchain ecosystems. This technology ensures that pre-defined conditions are met before executing an agreement, minimizing risks associated with fraud and human error [10]. One way to design smart contracts is by means of unified modeling language (UML). ...

Reference:

Multi-layer Track & Trace System Landscape with Smart Contract Validation Instances

... Unfortunately, the traditional data-driven soft sensing methods rely on the machine learning and deep learning technology, and therefore have black-box characteristics and lack interpretability. In recent years, the interpretability of soft sensors has received increasing attention [18], [19]. Some work hope to provide post hoc explanations for these black-box models. ...

Reference:

Bayesian-Based Causal Structure Inference With a Domain Knowledge Prior for Stable and Interpretable Soft Sensing

... Consequently, most conversational systems available in the mental health field adhere to a rule-based approach. [24] With MentalQuery we have addressed the challenge of combining these paradigms to achieve valid and accurate responses while also allowing for more flexible language understanding and successful handling of out of domain requests. As described in Section 3, we have combined three dialogue management approaches combining rules and LLMs in a way that is transparent to the user, applying each technique to the conversational context that best aligns with its respective strengths and limitations. ...

Reference:

MentalQuery: a proposal for conversational human-robot interaction to promote mental health literacy

... These questions, detailed in Table 2, were essential for establishing GADO's domain coverage and ensuring its capability to support clinical decision-making. Interestingly, these very same competency questions were also pivotal in directing the development process and later served as a metric for ontology verification [51]. ...

Reference:

A Domain Ontology for Gait Analysis and Decision Support on Gait-Related Diseases

... Furthermore, our study focused primarily on Spanish, which has its own set of challenges, particularly given the scarcity of Spanish-specific NLP datasets [19]. While we anticipate improvements in this regard with the recent development of Spanish corpora, such as the introduction of esCorpius [20], this remains a significant limitation of the current study. Also, the creation of new Spanish datasets, especially for distractor generation, could be a In addition, as previously mentioned, the use of the mT5 model and a machine-translated dataset likely contributed to the lower results for the question generation task, when compared to the monolingual T5 model. ...

Reference:

Automated multiple-choice question generation in Spanish using neural language models

... In [29], a framework based on consortia and smart contracts is introduced for monitoring workflow in agricultural food supply chains. This approach requires farmers to record their crops' environmental details and growth data in a centralized system that stores this information. ...

Reference:

Smart Contracts as a Tool to Support the Challenges of Buying and Selling Coffee Futures Contracts in Colombia

... Dengan demikian, BERT mampu digunakan untuk berbagai tugas bahasa termasuk analisis sentimen. Model ini menjadi memiliki keunggulan untuk menangkap hubungan konteks kata secara mendalam sehingga memberikan performa yang lebih akurat dibandingkan model bahasa terdahulu seperti Naïve Bayes (Fernández-Martínez et al., 2022). Zardak et al., (2023) melakukan studi analisis sentimen pada teks dokumen berbahasa Farsi dengan menggunakan ParsBERT. ...

Reference:

ANALISIS SENTIMEN PARA KANDIDAT PILPRES 2024 DENGAN MODEL BAHASA BERT

... Seven emotional classes, namely, angry, calm, disgust, fear, happy, sad, and surprise, were built from 1,440 videos of 24 professional actors (12 females and 12 males). We applied the hyperparameters of the original model, and because the five-fold CV setting of the Intermediate Attention fusion model is ambiguous, we borrowed the five-fold stratified CV setting from a recent paper (Luna-Jiménez et al., 2021). We included the validation set for training and used the early stopping with a patience of ten epochs (monitoring training loss as a criterion). ...

Reference:

Deep metric loss for multimodal learning

... This has led to it is application in various tasks, including sentiment and emotion recognition [24], [25], showcasing it is versatility across domains. Jiménez et al. [26] proposed an emotion recognition system utilizing transfer learning techniques. The framework involved a pre-trained spatial transformer network on saliency maps and facial images, followed by a bidirectional long short-term memory (BiLSTM) with an attention mechanism. ...

Reference:

Optimized multi-layer self-attention network for feature-level data fusion in emotion recognition