Knowledge Graph Question Answering (KGQA) aims to generate factual answers to natural language (NL) questions by querying structured data in knowledge graphs (KGs). This study investigates the use of in-context learning (ICL) for text-to-SPARQL generation, focusing on how the number, similarity, and diversity of prompt examples can affect an LLM' performance when applied to the text-to-SPARQL task. We conducted experiments using the LC-QuAD 1.0 benchmark and a 70B-parameter instruction-tuned LLM, exploring multiple dynamic example selection strategies based on BERT embeddings. We evaluated performance based on the retrieved answers and the structural similarity of the queries. We found that adding similar examples to the prompt enhanced performance; however, the gains became minimal beyond approximately ten examples. Contrary to our initial hypothesis, diversity did not yield the expected improvements. We also conducted an error analysis and further experiments, revealing a recurring issue: even when gold URIs are provided, the models often generate fake URIs, and additional problems persist.

Accurate classification of rare and undesirable events is crucial to maximizing the safety and efficiency of the oil and gas industry. While deep learning techniques have shown promising results in this domain, they often struggle when training data is scarce, a common issue with rare events. Domain ontologies can formally represent domain knowledge about these events. In this work, we hypothesize that it is possible to use domain knowledge represented in ontologies to improve the performance of neural networks when data is limited. Based on that, we propose a novel method by integrating domain knowledge from an ontology with deep learning through a hybrid loss function. Our hybrid loss function combines crossentropy with cosine distance between the ontological embeddings of the predicted and true classes to enforce semantic accuracy. Experiments conducted on the 3W Dataset, which includes nine rare anomalous events, demonstrate that our approach can improve classification accuracy compared to standard methods on smaller datasets.

While deep learning, including Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), has significantly advanced classification performance, its typical reliance on extensive annotated datasets presents a major obstacle in many practical scenarios where such data is scarce. Vision-language models (VLMs) and transfer learning with pre-trained visual models appear as promising techniques to deal with this problem. This paper proposes a novel zero-shot image classification framework that combines a VLM and a pre-trained vision model within a self-learning cycle. Requiring only the set of class names and no labeled training data, our method utilizes a confidence-based pseudo-labeling strategy to train a lightweight classifier directly on the test data, enabling dynamic adaptation. Our approach avoids VLM fine-tuning and the use of large language models, relying on the vision-only model to reduce the dependence on semantic representation. Experimental evaluations on ten diverse datasets demonstrate that our approach outperforms the baseline zero-shot method.

The growing adoption of machine learning techniques in recent years has led to an unprecedented demand for processing vast amounts of data, driving a tremendous increase in the consumption of computational resources and energy. Prototype selection techniques have been applied to reduce the requirements of computational resources and energy consumption that are needed by these techniques. In this paper, we propose an approach for speeding up existing prototype selection techniques. Firstly, it splits the dataset into a set of spatial partitions. Then, it builds a set of prototypes fo representing the information within each spatial partition. After, some conventional prototype selection algorithms can be applied to the prototypes selected by our approach. The experimental results demonstrate that, overall, our proposed approach maintains accuracy while enhancing the reduction rate of the original prototype selection algorithms and simultaneously reducing their execution times.

Recent advancements in Natural Language Processing (NLP) have highlighted the success of transformer-based models in various tasks, including Named Entity Recognition (NER). This study evaluates and compares these models using GeoCorpus-3, Portuguese's most extensive annotated geological corpus. We investigate the performance of BERTimbau and XLM-RoBERTa models across 30 entity classes, incorporating both linear and CRF layers, and assess their effectiveness using micro, macro, and weighted F1-scores. Our results reveal that transformer models, particularly XLM-RoBERTa, achieve superior performance compared to previous methods. Statistical analysis indicates significant performance differences, with BERTimbau-Linear showing lower performance than the other models. This study provides a comprehensive evaluation of transformer models on a geological corpus and makes the updated corpus and models publicly available to advance research and development in this domain.

Recent advancements in deep learning techniques, such as the introduction of convolutional neural networks (CNNs) and transformers, have led to significant improvements in the accuracy and efficiency of image classification tasks. Despite these advancements, deep learning methods generally require a large amount of annotated data, which can be challenging in domains where labeled data is scarce. Transfer learning strategies have become a promising solution to this issue. This study compares the performance of recent pre-trained neural networks for feature extraction in image classification tasks. Additionally, we analyze performance patterns across models to identify common trends in their classification accuracy and feature extraction capabilities across diverse datasets. We considered 13 different pre-trained models across six image datasets. Our findings indicate that MetaCLIP, EVA-02, and ViT_large achieved the best overall performance across the selected datasets. Furthermore, the experiments also indicate clusters of models that exhibit highly similar behaviors across the datasets. Our results can support model selection for feature extraction in other image classification tasks.

Domain-specific ontologies support system design and can establish a framework for fulfilling user-level, safety, or ethical requirements. The IEEE 7007-2021 Ontological Standard for ethically driven robotics and automation systems is the first industry standard to introduce a structure of ontologies concerning robot ethics and related fields, such as data privacy, transparency, responsibility, and accountability, offering a systems science approach to support the ethically aligned design of complex cyber–physical systems (CPSs) and robots particularly. This article provides a comprehensive overview of the main ontological commitments composing the foundation of the standard, the rationale behind their development, together with use cases of applications. Future directions for ethically aligned robotics and artificial intelligence (AI)-based systems along IEEE 7007-2021 are outlined, taking into account the exponentially growing fields of service and medical robotics.

Word semantic similarity is essential to numerous natural language processing and computational linguistics applications, including word sense disambiguation, machine translation, and information retrieval. Evaluating word similarity better and faster remains a complex problem due to the limitations of semantic similarity algorithms. In this paper, we propose a novel approach for computing word semantic similarity that uses the disjoint axioms between top-level ontology concepts as a heuristic for evaluating the semantic similarity between domain entities that represent the senses of those words. Our method leverages the disjointness information in top-level ontologies to reduce the number of comparisons needed to evaluate the semantic similarity between words. We evaluate our approach on benchmark datasets for the word similarity task. The results show that our method outperforms the traditional algorithms for knowledge-based word similarity regarding correlation with human judgments and time performance. Our proposed method can potentially facilitate the development of more effective and efficient applications.

Accurate production forecasting from oil well is an important decision-making variable for the Oil & Gas industry. Conventionally, future production is estimated using mechanistic modeling of the fluid through the reservoir rock and well flowlines. These models consider fluid properties, reservoir geometry, rock properties, subsurface conditions such as pressures, temperatures, and effects of flow-restricting valves and injection wells. This type of modeling is often computationally expensive since physical behavior is solved numerically. Furthermore, conventional methods rely on data that is costly to acquire or can hinder production due to the need for model calibration. Due to the advancements in subsurface sensor technology, a big amount of pressure and temperature data is gathered from wells in near real-time and at high frequencies. Data-driven methods are particularly well-suited for mapping this sensor and operational data into oil flow forecasting. In this context, we compared the performance of different state-of-the-art data-driven models for forecasting transient oil flow that are trained and tested on hourly well data obtained from a real-world petroleum operation. Additionally, we investigated how adjusting the lookback and horizon sizes for the training samples influences the model’s performance and explored the impact of future operational variables on oil rate forecasting. Results showed that ensemble tree-based models performed better than deep learning techniques for this particular set of data, probably due to the restricted size and variability of the dataset. Lookback and horizon sizes play a role in forecast accuracy depending on the specific model. We also determined that incorporating future operational features is useful for improving forecasting accuracy. We conclude that better standards for this specific task and better pre-processing and feature selection practices need to be further evaluated.

Due to the increasing size of the datasets, instance selection techniques have been applied for reducing the computational resources involved in data mining and machine learning tasks. In this paper, we propose an attraction-based approach for selecting instances. The algorithm adopts the notion of attraction for selecting the most representative instances of each class. The resulting approach allows the user to define how many representative instances should be selected. Our method was evaluated in a classification task considering 14 well-known datasets. The performance of the proposed algorithm was compared to the performances of 8 prototype selection algorithms in terms of accuracy and reduction rate. The experimental results show that, in general, the proposed algorithm provides a good trade-off between reduction rate and accuracy with a low time complexity.

In this paper, we propose a soft subspace hierarchical clustering for dealing with categorical data. The proposed algorithm extends the traditional agglomerative hierarchical clustering approach for identifying clusters of categorical data in subspaces. The algorithm adopts a correlation-based approach for measuring the relevance of each categorical attribute during the clustering process. We performed experiments on six well-known datasets, comparing the performance of our algorithms with the original agglomerative algorithm for hierarchical clustering and other five partitional subspace clustering algorithms, using two well-known evaluation metrics: accuracy and f-measure. According to the experiments, the proposed algorithm outperforms the original one. Besides that, the proposed algorithm outperforms most of the partitional algorithms, while provides additional advantages.

In the context of natural language processing and information retrieval, ontologies can improve the results of the word sense disambiguation (WSD) techniques. By making explicit the semantics of the term, ontology-based semantic measures play a crucial role to determine how different ontology classes have a similar meaning. In this context, it is common to use semantic similarity as a basis for WSD. However, the measures generally consider only taxonomic relationships, which affects negatively the discrimination of two ontology classes that are related by the other relationship types. On the other hand, semantic relatedness measures consider diverse types of relationships to determine how much two classes on the ontology are related. However, these measures, especially the path-based approaches, has as the main drawback a high computational complexity to be calculated in query execution time. Also, for both types of semantic measures, it is unpractical to store all similarity or relatedness values between all ontology classes in memory, especially for large ontologies. In this work, we propose a novel approach based on semantic neighbors that aim to improve the query time in path-based semantic measures without losing their effectiveness in relatedness analysis. We also propose an efficient algorithm to calculate the semantic distance between two ontology classes. We evaluate our proposal in WSD using a pre-existent domain ontology for well-core description. This ontology contains 929 classes related to rock facies and a set of sentences from four different corpora about geology in the Oil&Gas domain. In the experiments, we compared our approach with state-of-the-art semantic relatedness measures, such as path-based, feature-based, information content, and hybrid methods regarding the F-score, query time and the total number of classes in memory. The experimental results show that the proposed method obtains F-score gains in WSD, as well as an improvement in the query time concerning the traditional path-based approaches. Also, we reduce the total number of classes stored in memory for each ontology class.

Machine learning approaches have been applied in huge volumes of data. In order to deal with this big data, techniques for instance selection have been applied for reducing the data to a manageable volume and, consequently, for reducing the computational resources that are necessary to apply machine learning approaches. In this paper, we propose an efficient approach for instance selection called ISDSP. It adopts the notion of spatial partition for efficiently splitting the dataset in sets of similar instances. In a second step, the algorithm selects a representative instance of each of the densest spatial partitions that were previously identified. The approach was evaluated on 15 well-known datasets used in a classification task, and its performance was compared to those of 6 state-of-the-art algorithms, considering two measures: accuracy and reduction. All the obtained results show that, in general, the proposed approach provides a good trade-off between accuracy and reduction, with a significantly lower running time, when compared with other approaches.

Nowadays, data mining approaches have been applied for extracting useful knowledge from a huge volume of data. In order to deal with this big data, techniques for prototype (or instance) selection have been applied for reducing the data to a manageable volume and, consequently, for reducing the computational resources that are necessary to apply data mining approaches. However, most of the proposed approaches for prototype selection have a high time complexity and, due to this, they cannot be applied for dealing with big data. In this paper, we propose an efficient approach for prototype selection called PSSP. It adopts the notion of subspace partition for efficiently splitting the dataset in sets of similar instances. In a second step, the algorithm extracts a prototype of each of the previously identified sets. The approach was evaluated on 11 well-known datasets used in a classification task, and its performance was compared to those of 6 state-of-the-art algorithms, considering three measures: accuracy, reduction, and effectiveness. All the obtained results show that, in general, the proposed approach provides a good trade-off between accuracy and reduction, with a significantly lower running time, when compared with other approaches.

Creating a standard for knowledge representation and reasoning in autonomous robotics is an urgent task if we consider recent advances in robotics as well as predictions about the insertion of robots in human daily life. Indeed, this will impact the way information is exchanged between multiple robots or between robots and humans and how they can all understand it without ambiguity. Indeed, Human Robot Interaction (HRI) represents the interaction of at least two cognition models (Human and Robot). Such interaction informs task composition, task assignment, communication, cooperation and coordination in a dynamic environment, requiring a flexible representation. Hence, this paper presents the IEEE RAS Autonomous Robotics (AuR) Study Group, which is a spin-off of the IEEE Ontologies for Robotics and Automation (ORA) Working Group, and and its ongoing work to develop the first IEEE-RAS ontology standard for autonomous robotics. In particular, this paper reports on the current version of the ontology for autonomous robotics as well as on its first implementation successfully validated for a human-robot interaction scenario, demonstrating the developed ontology's strengths which include semantic interoperability and capability to relate ontologies from different fields for knowledge sharing and interactions.

Due to the increasing of the size of the datasets, techniques for instance selection have been applied for reducing the data to a manageable volume, leading to a reduction of the computational resources that are necessary for performing the learning process. Besides that, algorithms of instance selection can also be applied for removing useless, erroneous or noisy instances, before applying learning algorithms. In the last years, several approaches for instance selection have been proposed. However, most of them have high time complexity and, due to this, they cannot be used for dealing with large datasets. In this paper, we present an algorithm called CDIS that can be viewed as an improvement of a recently proposed density-based approach for instance selection. The main contribution of this paper is a formal characterization of a novel density function that is adopted by the CDIS algorithm. The CDIS algorithm evaluates the instances of each class separately and keeps only the densest instances in a given (arbitrary) neighborhood. This ensures a reasonably low time complexity. Our approach was evaluated on 20 well-known data sets and its performance was compared with the performance of 6 state-of-the-art algorithms, considering three measures: accuracy, reduction and effectiveness. For evaluating the accuracy achieved using the datasets produced by the algorithms, we applied the KNN algorithm. The results show that our approach achieves a performance (in terms of balance of accuracy and reduction) that is better or comparable to the performances of the other algorithms considered in the evaluation.

Instance selection is an important preprocessing step that can be applied in many machine learning tasks. Due to the increasing of the size of the datasets, techniques for instance selection have been applied for reducing the data to a manageable volume, leading to a reduction of the computational resources that are necessary for performing the learning process. Besides that, algorithms of instance selection can also be applied for removing useless, erroneous or noisy instances, before applying learning algorithms. This step can improve the accuracy in classification problems. In the last years, several approaches for instance selection have been proposed. However, most of them have long runtimes and, due to this, they cannot be used for dealing with large datasets. In this paper, we propose a simple and effective density-based approach for instance selection. Our approach, called LDIS (local density-based instance selection), evaluates the instances of each class separately and keeps only the densest instances in a given (arbitrary) neighborhood. This ensures a reasonably low time complexity. Our approach was evaluated on 15 well-known data sets and its performance was compared with the performance of 5 state-of-the-art algorithms, considering three measures: accuracy, reduction and effectiveness. For evaluating the accuracy achieved using the datasets produced by the algorithms, we applied the KNN algorithm. The results show that LDIS achieves a performance (in terms of balance of accuracy and reduction) that is better or comparable to the performances of the other algorithms considered in the evaluation.

The interest in attribute weighting, in clustering tasks, have been increasing in the last years. However, few attempts have been made to apply automated attribute weighting to categorical data clustering. Most of the existing approaches computes the weights based on the frequency of the mode category or according to the average distance of data objects from the mode of a cluster. In this paper, we adopt a different approach, investigating how to use the correlation among categorical attributes for measuring their relevancies in clustering tasks. As a result, we propose a correlation-based attribute weighting approach for categorical attributes.

The interest in attribute weighting for soft subspace clustering have been increasing in the last years. However, most of the proposed approaches are designed for dealing only with numeric data. In this paper, our focus is on soft subspace clustering for categorical data. In soft subspace clustering, the attribute weighting approach plays a crucial role. Due to this, we propose an entropy-based approach for measuring the relevance of each categorical attribute in each cluster. Besides that, we propose the EBK-modes (entropy-based k-modes), an extension of the basic k-modes that uses our approach for attribute weighting. We performed experiments on five real-world datasets, comparing the performance of our algorithms with four state-of-the-art algorithms, using three well-known evaluation metrics: accuracy, f-measure and adjusted Rand index. According to the experiments, the EBK-modes outperforms the algorithms that were considered in the evaluation, regarding the considered metrics.

In this paper we present an automatic approach for the litho logic correlation task, which is a crucial task in the domain of Petroleum Geology. In our approach, we consider the litho logic correlation task as formally equivalent to the task of DNA sequence alignment, in the domain of Bioinformatics. This allows us to use well-known sequence alignment algorithms for dealing with this task. Moreover, our approach uses a domain ontology for representing the knowledge about the domain objects. This ensures an agreement regarding the formal vocabulary used for describing the geological objects, allowing us to deal with the data collected by different geologists in a homogeneous way. Finally, we also use clustering techniques for finding groups of geological objects that are sufficiently similar for being correlated by the sequence alignment algorithms. We discuss some results of the application of our approach to a real dataset.

Ontologies have been applied in Computer Science to ensure the semantic interoperability among multiple systems. With the increasing of ontologies availability, many approaches for promoting the share and reuse of ontologies have been investigated in recent years, like ontology module extraction (modularization) and ontology view extraction. Approaches for ontology module extraction are used for extracting modules in large ontologies. On the other hand, ontology views have been used for providing to the user only the parts of the ontology that are useful for a given task. Thus, both ontology views and ontology modules encapsulate a subset of the original ontology, but they have different purposes. The literature has explored the ontological meta-properties (such as identity and rigidity) for guiding the modeling decisions that are made during the ontology engineering process. Some of these meta-properties were formalized in foundational ontologies like UFO (Unified foundational ontology). In this paper, we explore the use of ontological meta-properties for extracting ontology views. We propose a characterization of the notion of well-founded ontology view, considering the ontological meta-properties of the concepts. Besides that, we also propose a language-independent algorithm for sub-ontology extraction that is guided by ontological meta-properties. Finally, we present a case study for illustrating the application of our approach.

In visual domains, the tasks are accomplished through intensive use of visual knowledge. In this paper, we are interested in the visual interpretation task, which is prevalent in many visual domains. We investigate the role played by foundational ontologies in reasoning processes that deals with visual information, as those that are performed in visual interpretation tasks. We propose an approach for visual interpretation that combines ontologically well founded domain ontologies, a reasoning model, and a cognitively well founded meta-model for representation of inferential knowledge. Our approach was effectively applied in the task of visual interpretation of depositional processes, within the Sedimentary Stratigraphy domain.

Unambiguous definition of spatial position and orientation has crucial importance for robotics. In this paper we propose an ontology about positioning. It is part of a more extensive core ontology being developed by the IEEE RAS Working Group on ontologies for robotics and automation. The core ontology should provide a common ground for further ontology development in the field. We give a brief overview of concepts in the core ontology and then describe an integrated approach for representing quantitative and qualitative position information.