The rapid growth of online services has heightened concerns about user protection from cyber threats, particularly phishing, which poses significant risks to cyber-social security. To this end, we propose a novel tool for phishing detection called U-Proof. Our tool uses both state-of-the-art LLMs and traditional ML models to detect phishing websites. In particular, we evaluate the phishing detection capabilities of different LLMs and compare them with several ML models to analyze the impact of different model architectures on the identification of phishing websites. For a comprehensive experimental evaluation, we use a combination of public and custom datasets. These include active phishing websites from September 2024, as well as URLs from banks and postal services. Furthermore, the tool includes explanations to enhance user awareness of phishing tactics, supporting broader educational efforts to reduce risks.

Household energy saving has become increasingly important, and finding Machine Learning models that allow residents to predict their consumption, particularly when equipped with a renewable energy source, such as photovoltaic (PV) panels, is a central challenge in achieving the goal of greater energy efficiency. Several models and approaches have been currently used to address this challenge. In this paper we present a series of foundational experiments that provide valuable insights into the task of net load forecasting using some well-known models available for regression tasks, i.e. the Multi-Layer Perceptron, CatBoost and XGBoost. We tested these models with the specific aim of assessing their robustness in a scenario where a decreasing time frame is used for their training, and resorting to a household equipped with PV plant and a battery storage system as use case. The results, measured by R2score, show notable performance for both MLP and XGBoost models, even in data-scarce settings, while CatBoost lags behind, proving to be less effective.

Now in its third edition, the Data Exploration in the Web 3.0 Age (DEW) track of the IEEE International WETICE Conference continues to bring together researchers and practitioners from both the Academia and Industry working in the areas related to data exploration, in a very broad sense. Papers accepted for presentation at DEW 2019 are representatives of emerging topics in the fields of data and text mining, machine learning, semantic web and Internet of things.

We present a Question Answering system aimed to answer natural language questions over the open RDF spending data provided by LinkedSpeding. We propose an original machine-learning approach to learn generalized SPARQL templates from an existing training set of (NL question, SPARQL query) pairs. In our approach, the generalized SPARQL templates are fed to an instance-based classifier that associates a given user-provided question to an existing pair that is used to answer the user question. We employ an external tagger, delegating the Named-Entity Recognition (NER) task to a service developed for the domain we want to query. The problem is particularly challenging due to the small training set size available, counting only 100 questions/SPARQL queries. We illustrate the results of our new approach using data provided by the Question Answering over Linked Data challenge (QALD-6) task 3, showing that we can provide a correct answer to 14 of the 50 questions of the test set. These results are then compared to existing systems, including our previous system, QA3, where templates were provided by an expert rather than being generated automatically from a training set.

Data Exploration in the Web 3.0 Age (DEW) is a track of the IEEE International Conference on Enabling Technologies: Infrastructure for Collaborative Enterprises (WETICE). DEW aims at bringing together researchers and practitioners from both the Academia and Industry working in the areas related to data exploration, in a very broad sense. The track welcomes contributions from data mining, query languages, data visualization, graph databases and other fields related to the analysis and exploitation of data. This summary report includes a brief description of the papers accepted for presentation at the 2018 edition of DEW.

In this paper, we leverage advances in the Semantic Web area, including data modeling (RDF), data management and querying (JENA and SPARQL), to develop CodeOntology , a community-shared software framework supporting expressive queries over source code. The project consists of two main contributions: an ontology that provides a formal representation of object-oriented programming languages, and a parser that is able to analyze Java source code and serialize it into RDF triples. The parser has been successfully applied to the source code of OpenJDK 8, gathering a structured dataset consisting of more than 2 million RDF triples. CodeOntology allows to generate Linked Data from any Java project, thereby enabling the execution of highly expressive queries over source code, by means of a powerful language like SPARQL.

The goal of the Data Exploration in the Web 3.0 Age (DEW 2017) conference track has been bringing together researchers and practitioners from both the Academia and Industry working in the areas related to data exploration, in a very broad sense. The track welcomes contributions from data mining, query languages, data visualization, graph databases and other fields related to the analysis and exploitation of data. This paper is a Summary Report for the Data Exploration in the Web 3.0 Age (DEW) Track that includes the list of accepted papers.

Despite the significant contribution from specialized ontologies and text mining methods, the evaluation of the semantic similarity of genes remains difficult because of the complex functions in which genes are involved. A less exploited resource is Wikipedia that stores more than 10400 articles about human genes: each gene name identifies the corresponding Wikipedia page resuming gene's properties in short sentences where hyperlinks define relationships with other genes in Wikipedia. This paper evaluates the extent to which the Wikipedia can be trusted for assessing the similarity of a gene pair as the distance between their Wikipedia pages. We present a set of experiments that make use of TagMe (a powerful tool for evaluating the distance of two Wikipedia pages based on their annotations) to calculate the semantic similarity of several sets of genes on Wikipedia. Results compare well with gold standards and semantic similarity values evaluated on gene ontologies. The paper demonstrates the effectiveness of Wikipedia in recognizing functional groups of genes, the quality and the wealth of its knowledge about genes as well the accuracy of TagMe.

This paper discusses expressivity and accuracy of the By-Example Structured (BESt) Query paradigm implemented on the SWiPE system through the Wikipedia interface. We define an experimental setting based on the natural language questions made available by the QALD-4 challenge, in which we compare SWiPE against Xser, a state-of-the-art Question Answering system, and plain keyword search provided by the Wikipedia Search Engine. The experiments show that SWiPE outperforms the results provided by Wikipedia, and it also performs sensibly better than Xser, obtaining an overall 85% of totally correct answers vs. 68% of Xser. Among all answered questions, we obtain a precision of 100% and recall 96%. SWiPE is also able to answer more questions than the other systems. A formal characterization of the set of SPARQL queries supported by the BESt Query paradigm is also provided.

In many Semantic Web applications, having RDF predicates sorted by significance is of primarily importance to improve usability and performance. In this paper we focus on predicates available on DBpedia, the most important Semantic Web source of data counting 470 million english triples. Although there is plenty of work in literature dealing with ranking entities or RDF query results, none of them seem to specifically address the problem of computing predicate rank. We address the problem by associating to each DBpedia property (also known as predicates or attributes of RDF triples) 8 original features specifically designed to provide sort-by-importance quantitative measures, automatically computable from an online SPARQL endpoint or a RDF dataset. By computing those features on a number of entity properties, we created a learning set and tested the performance of a number of well-known learning-to-rank algorithms. Our first experimental results show that the approach is effective and fast. Further, we provide an extensive survey of state-of-the-art algorithms for RDF ranking, to which we compare our approach.

Spatio-temporal, geo-referenced datasets are growing rapidly, and will be more in the near future. This phenomenon is mostly due to the daily collection of telecommunication data from mobile phones and other location-aware devices and is expected to enable novel classes of applications based on the extraction of behavioral patterns from mobility data. Such patterns could be used for instance in traffic and sustainable mobility management (e.g., to study the accessibility to services), urban planning, environmental monitoring, and collaborative location-based services. Clearly, in these applications privacy is a concern, since some knowledge may be sensitive, or an over-specific pattern may reveal the behaviour of groups of few individual. In this paper we focus on automated privacy-preserving methods we developed for extracting and sharing user- consumable forms of knowledge from large amounts of raw data referenced in space and in time.

Advances in information technology, and its use in research, are increasing both the need for anonymized data and the risks of poor anonymization. We present a metric, δ-presence, that clearly links the quality of anonymization to the risk posed by inadequate anonymization. We show that existing anonymization techniques are inappropriate for situations where δ-presence is a good metric (specifically, where knowing an individual is in the database poses a privacy risk), and present algorithms for effectively anonymizing to meet δ-presence. The algorithms are evaluated in the context of a real-world scenario, demonstrating practical applicability of the approach.

Privacy-preserving data mining is an important area that studies privacy issues of data mining. When the goal is to share data mining results, two privacy-related problems may arise. The first one is how to compute the data-mining results among several parties without sharing the data. Cryptography-based primitives are the basic tool used to develop ad-hoc secure multi-party computation protocols that share information as less as possible during the computation under different adversary models. The second one is how to produce data mining results that provably do not contain threats to the anonymity of individuals. The concept of k-anonymity has been used to discover anonymity-preserving frequent patterns, and centralized algorithms have been developed. In this paper and for the first time, we study how to produce anonymity-preserving data mining results in a distributed environment. We present two privacy-preserving strategies and show their feasibility through experimental analysis.

In this paper we study when the disclosure of data mining results represents, per se, a threat to the anonymity of the individuals recorded in the analyzed database. The novelty of our approach is that we focus on an objective definition of privacy compliance of patterns without any reference to a preconceived knowledge of what is sensitive and what is not, on the basis of the rather intuitive and realistic constraint that the anonymity of individuals should be guaranteed. In particular, the problem addressed here arises from the possibility of inferring from the output of frequent itemset mining (i.e., a set of item-sets with support larger than a threshold a), the existence of patterns with very low support (smaller than an anonymity threshold k)[M. Atzori et. al, 2005]. In the following we develop a simple methodology to block such inference opportunities by introducing distortion on the dangerous patterns.