This paper presents a dataset of Lithuanian language comments annotated with emotional manipulation techniques. The source material comprises comments from Lithuanian news portal texts. A total of 1000 comments were selected and manually annotated by four human annotators using Label Studio. The annotators identified text fragments corresponding to fourteen emotional manipulation techniques, producing span-level human annotations that form the primary component of the dataset.In addition to the manual annotations, the dataset includes machine-generated outputs created by the GPT-4.1 model. These outputs were produced by executing prompts designed for the detection and classification of emotional manipulation span. Several prompting strategies were applied, and each prompt was run five times to capture variability in model behaviour. Because GPT-4.1 does not always return verbatim text excerpts, all generated spans were post-processed to extract precise fragments from the original comments and compute their character-level offsets.The resulting dataset encompasses four components: unprocessed source comments, human-annotated spans, prompt templates, and GPT-4.1 generated predictions.The dataset provides the first publicly available Lithuanian resource annotated for emotional manipulation techniques and can support research on manipulation and persuasion phenomena in a morphologically rich, low-resource language. The human annotations may be used as a benchmark for evaluating computational models for span extraction and technique classification. The inclusion of prompts and corresponding GPT-4.1 outputs enables the study of large language model behaviour under different prompting strategies and facilitates prompt engineering research without repeating inference runs. The dataset may also be reused for training or evaluating multilingual and cross-lingual models and offers a foundational framework for the development of annotation schemes and guidelines in related corpus construction efforts.

This paper examines whether incremental prompt engineering can enable reliable large language model (LLM)-based pre-annotation of corpus texts in a low-resource language setting. Using Lithuanian as a case study, we systematically evaluate multiple LLM prompt designs and assess their suitability for generating emotional manipulation annotations for corpus development. We find that performance varies with task complexity, and systematic prompt refinement measurably reduces output instability. Cross-model evaluation of the best-performing prompting strategy shows consistent and similar trends over several modern LLMs. Our results demonstrate that while structured prompts substantially improve output consistency and LLM-assisted annotation can roughly approximate human-produced labels for well-defined categories, the quality of results produced by contemporary LLMs is unsatisfactory for automatic pre-annotation of emotional manipulation techniques in a low-resource language.

We present a system to improve the privacy and security of personally identifiable customer data within e-commerce. This system utilizes a trusted mediator to store authentic customer data, while substituting it with synthetic data in the systems of e-commerce participants. Our system is designed to be retrofittable onto existing e-commerce platforms.

Online hate speech and abusive language pose a growing challenge for content moderation, especially in multilingual settings and for low-resource languages such as Lithuanian. This paper investigates to what extent modern multilingual sentence embedding models can support accurate hate speech detection in Lithuanian, Russian, and English, and how their performance depends on downstream modeling choices and feature dimensionality. We introduce LtHate, a new Lithuanian hate speech corpus derived from news portals and social networks, and benchmark six modern multilingual encoders (gemma, qwen, bge, snow, jina, and e5) on LtHate, RuToxic, and EnSuperset using a unified Python pipeline. For each embedding type, we train both a one-class histogram-based anomaly detector (HBOS) and a two-class gradient-boosted tree ensemble (CatBoost), with and without Principal Component Analysis (PCA) compression to 32-dimensional feature vectors. Across all datasets, two-class supervised models consistently and substantially outperform one-class anomaly detection, with the best configurations achieving up to 78.8% accuracy (Kappa 0.58, AUC ROC 0.87) in Lithuanian (jina), 92.2% accuracy (Kappa 0.77, AUC ROC 0.97) in Russian (e5), and 76.9% accuracy (Kappa 0.48, AUC ROC 0.86) in English (e5). PCA compression deteriorates the discriminative power of CatBoost only slightly, with much more negative impact for the HBOS model. These results demonstrate how modern multilingual sentence embeddings combined with gradient-boosted decision trees provides robust machine learning solutions for multilingual hate speech detection applications.

The performance characteristics and certain maintenance aspects of a natural language querying (NLQ) interface depend on how the data model is conceptualized. One of the areas where alternative conceptualizations are available is events and their specialization. The concept of event specialization is already known from event extraction methods, which allows for a more precise description of the events identified in a text. In the context of NLQ interfaces, event specialization allows narrower or broader questions. This study investigates how the choice of event specialization strategy in OWL (Web Ontology Language) ontologies affects the complexity and performance-related aspects of the NLQ interface to ontologies. In this paper, we present four event specialization strategies and investigate how they impact the size of the ontology schema and vocabulary of the NLQ interface, the performance of querying and data import, the size of the semantic repository, and the complexity of SPARQL queries. We discuss the strengths and weaknesses of each approach and present recommendations on determining the best one for the needs of NLQ interface end-users and developers.

Inadequate early scope estimation is a common problem in software projects, leading to failures in meeting project requirements. Agile projects usually do not concentrate on a comprehensive requirements analysis and specification before the start of the project, making scope assessment difficult. This paper presents the methodology for facilitating a more accurate early estimation of project scope, based on requirements information gathered in various forms (requirements models and textual descriptions) during the requirements workshop. The requirements from different sources are compiled into one list and reconciled, since they are prepared by a number of participants in the requirements workshop using different notations (UML diagrams, SysML models, Story map) and may have differences in the vocabulary. Reconciliation encompasses the unification of vocabulary, as well as the identification and the removal of overlaps in requirements. The final list of requirements is used to estimate the scope of the project in story points. The estimate can be presented to the client and used as a basis for the project contract. A case study on the application of the proposed methodology is presented, using the animal shelter information system as a development project. It demonstrates that the methodology is viable and can facilitate the gathering of a more extensive set of requirements, thus ensuring a more detailed scope estimation.

This study introduces a novel performance-based weighting scheme for ensemble learning using the Shapley value. The weighting uses the reciprocal of binary cross-entropy as a base learner's performance metric and estimates its Shapley value to measure the overall contribution of a learner to an equally weighted ensemble of various sizes. Two variants of this strategy were empirically compared with a single monolith model and other static weighting strategies using two large banking-related datasets. A variant that discards learners with a negative Shapley value was ranked as first or at least second when constructing homogeneous ensembles, whereas for heterogeneous ensembles this strategy resulted in a better or at least similar detection performance to other weighting strategies tested. The main limitation being the computational complexity of Shapley calculations, the explored weighting strategy could be considered as a generalization of performance-based weighting.

Open online courses are often used in formal education to provide added value for the students by helping gain new skills and competences or as extra material tasks in addition to the formal education course. The course described in this paper is developed and piloted fully open online. It is also integrated into formal education. This paper presents a case on integration of open databases course into formal education together with student feedback on the course quality and effectiveness of course delivery process.

Information in knowledge graphs is ordinarily accessed via queries in formal languages. Formal languages, however, present an obstacle for non-technical users. To make semantic search more convenient, it is desirable to enable use of natural language queries. One of the difficulties encountered in developing natural language querying systems is a mismatch between the way how users express their questions in a natural language and the way how a knowledge graph is structured. Such a mismatch is called semantic gap. One of the solutions for bridging the semantic gap is to use SBVR vocabularies for translation between the natural and formal languages. In this article, we investigate how alternative variants of event representation schemas affect properties of natural language querying interface: the size of ontology schema and vocabulary, the performance of querying and data import, repository size and query complexity.

SBVR is the richest knowledge model allowing to create specifications that are understandable for business people and also interpretable by computers. Existing SBVR editors still lack capabilities that could allow generating formal SBVR models, adapting SBVR to several languages or making SBVR extensions for various purposes without changing SBVR metamodel.The goal of the paper is to present the Xtext grammar for SBVR structured language and the prototype of SBVR editor, created on the base of this grammar. The experiment conducted with the prototype has shown that it allows defining business vocabularies, business rules and questions in SBVR structured English and Lithuanian languages; produxing formal SBVR models; using concepts from several vocabularies , and extending SBVR without changing its metamodel. DOI: http://dx.doi.org/10.5755/j01.itc.45.1.9219

The goal of the paper is to analyse the relevance of the proposed transformation from SBVR business vocabularies and business rules to the corresponding OWL 2 ontologies. The transformation was aimed to be model driven, lossless and reversible; covering transformable SBVR concepts and as much as possible OWL 2 concepts, especially those that are important for inference; supporting the consistency and normalization of resulting ontologies. The paper presents the detailed comparison of the proposed solution with the most advanced, to our knowledge, related works and experimental investigation of the implemented transformation prototype with nine SBVR business vocabularies and business rule sets with regards to defined desirable criteria. DOI: http://dx.doi.org/10.5755/j01.itc.45.2.8873

The goal of the paper is to analyse the subset of Semantics of Business Vocabulary and Business Rules (SBVR) for a comprehensive representation of ontological knowledge defined using the Web Ontology Language OWL 2. SBVR is the OMG metamodel, which separates the representation and meaning of business concepts and business rules, and makes them understandable for business experts as well as for software systems. The SBVR can act as an interface between business participants and semantic technologies, such as OWL 2 that has developed means for describing ontological data and reasoning with them. SBVR provides the richer model for knowledge representation than OWL 2. Though there are a few proposals that have shown that it is possible to transform the significant subset of SBVR concepts into OWL 2 ontology, the suitability of SBVR to represent OWL 2 ontologies has not been studied in detail. The paper addresses the mentioned issue with regards to the transformation from SBVR into OWL 2. DOI: http://dx.doi.org/10.5755/j01.itc.43.3.6651

We present a language independent method for detecting model implementation patterns in a source code. In contrast to most other reverse engineering methods, we exploit existing program model for this purpose. Our method works by recognizing instances of simple model-to-code trans format ions. The patterns we use for recognition of model elements can be reused for composing templates for generating a program code. Our method is used for recognizing relationships between program model and handwritten program code.

The OMG SBVR standard is the most mature abstract representation for Business semantics; however, the complexity of SBVR metamodel prevents its broad and rapid usage in business communities. There are a few SBVR implementations yet, but the popularity of SBVR is growing as different interest-groups are finding the variety of ways for applying the SBVR standard. The goal of this paper is to present the VeTIS tool capable for editing SBVR Business Vocabularies and Business Rules and transforming them into UML class models supplemented with OCL constraints. It is integrated into MagicDraw UML CASE tool and organically combines with simple development process, which pursues defining requirements via use cases and modelling business processes via activities.

In model-driven engineering (MDE) program code generators are used to generate program code from abstract program models, thus bringing the final code closer to program specification and saving time that would be spent in coding. Current approach to program code generation from abstract program models does not work well in quickly evolving environments due to the large amount of work that is required to fully prepare and maintain program code generator. This chapter presents analysis of current approach to program code generation and presents an alternative approach tailored for generating program code in quickly evolving environments by using self-configuring program code generator.

The most recent version of UML 2.0 (in year 2006) specifies activities and actions, which allow describing low level behavior of the system being modeled in implementation independent fashion. This paper represents general requirements for generating code from UML 2.0 activities and actions and proposes a way and requirements for expressing UML 2.0 activities with concurrent action execution in sequential program code.