Abstract Discovering relevant datasets in large, heterogeneous data ecosystems, such as Data Lakes or Data spaces, is a complex task, often hindered by a lack of transparency and user-centric explanations in the discovery process. Explainability is critical for enabling users to understand why specific datasets are recommended, what information they contain, and how they align with user-defined criteria and preferences. To address these challenges, this work proposes a novel Graph Retrieval-Augmented Generation (Graph RAG) framework to enhance explainability in a platform for discovery of summary data sources. The proposed approach leverages a Knowledge Graph (KG) to interpret user requests, extracting relevant contextual information. These enriched requests are then transformed by a Large Language Model (LLM) into actionable dataset queries for a dataset discovery platform. Candidate solutions are evaluated and enriched with statistical insights on value distributions and contextual knowledge from the KG. Finally, the LLM ranks these solutions based on user preferences, producing a final report. This dual strategy of query enrichment and contextual explanation fosters transparency and enhances user understanding of the discovery process. We demonstrate the effectiveness of the approach through an experimental validation, highlighting its potential to improve both the accuracy and interpretability of dataset discovery.

Model repair techniques aim at automatically updating a process model to incorporate behaviors that are observed in reality but are not compliant with the original model. Most state-of-the-art techniques focus on the fitness of the repaired models, with the goal of including single anomalous behaviors observed in a log in the form of the events. This often hampers the precision of the obtained models, which end up allowing much more behaviors than intended. In the quest of techniques avoiding this over-generalization pitfall, some notion of higher-level anomalous structure is taken into account. The type of structure considered is however typically limited to sequences of low-level events. In this work, we introduce a novel repair approach targeting more general high-level anomalous structures. To do this, we exploit instance graph representations of anomalous behaviors, that can be derived from the event log and the original process model. Our experiments show that considering high-level anomalies allows to generate repaired models that incorporate the behaviors of interest while maintaining precision and simplicity closer to the original model.

Predictive monitoring of business processes aims at predicting the future of an ongoing process execution. In this work, we focus on the prediction of the next activities to be executed in a running case. However, in contrast with most state-of-the-art approaches, focused on predicting exactly the next activity that will be executed from the current state of the process, we propose an approach aimed at predicting the portion of the process (or “location”) that is likely to be executed next. The notion of location allows us to detect activities belonging to the same portion of a control-flow construct (e.g., at the beginning of a parallelism, or at the end of a loop). It provides an abstraction mechanism from the level of the single activity, which can be used to provide the process analyst with an higher-level overview of what can be expected next in the process execution. We validated the approach over a set of real-world datasets comparing and discussing different strategies for training a classifier in returning a location in place of an activity label.

Data Lake (DL) architectures have recently emerged as an effective solution to the problem of data analytics with big, highly heterogeneous, and quickly changing data sources. However, novel challenges arise too, including how to make sense of disparate raw data and how to identify the sources that satisfy a data need. In the paper, we introduce a semantic model for a Data Lake aimed to support data discovery and integration in data analytics scenarios. By formally modeling indicators of interest, their computation formulas, and dimensions of analysis in a knowledge graph, and by seamlessly mapping them to relevant source metadata, the framework is suited for identifying the sources and the required transformation steps according to the analytical request.

Conformance checking allows organizations to compare process executions recorded by the IT system against a process model representing the normative behavior. Most of the existing techniques, however, are only able to pinpoint where individual process executions deviate from the normative behavior, without considering neither possible correlations among occurred deviations nor their frequency. Moreover, the actual control-flow of the process is not taken into account in the analysis. Neglecting possible parallelisms among process activities can lead to inaccurate diagnostics; it also poses some challenges in interpreting the results, since deviations occurring in parallel behaviors are often instantiated in different sequential behaviors in different traces. In this work, we present an approach to extract anomalous frequent patterns from historical logging data. The extracted patterns can exhibit parallel behaviors and correlate recurrent deviations that have occurred in possibly different portions of the process, thus providing analysts with a valuable aid for investigating nonconforming behaviors. Our approach has been implemented as a plug-in of the ESub tool and evaluated using both synthetic and real-life logs.

The class-imbalance problem is the problem of learning a classification rule from data that are skewed in favor of one class. On these datasets traditional learning techniques tend to overlook the less numerous class, at the advantage of the majority class. However, the minority class is often the most interesting one for the task at hand. For this reason, the class-imbalance problem has received increasing attention in the last few years. In the present paper we point the attention of the reader to a learning algorithm for the minimization of the average misclassification risk. In contrast to some popular class-imbalance learning methods, this method has its roots in statistical decision theory. A particular interesting characteristic is that when class distributions are unknown, the method can work by resorting to stochastic gradient algorithm. We study the behavior of this algorithm on imbalanced datasets, demonstrating that this principled approach allows to obtain better classification performances compared to the principal methods proposed in the literature.

In the information system view, the reference architecture for strategic and decision support is based on the Data Warehouse architecture, that enables flexible and multidimensional analysis of strategic indexes by means of OLAP tools and reports. In this paper we propose a novel model for semantic annotation of Data Warehouse schema that takes into account domain ontologies as well as a mathematical ontology. Such an ontology describes mathematical formulas underlying elements of the datacube schema, including the semantics of operands and operators. In particular, we discuss and apply the proposed model for the semantic annotation of the schema of a datacube, that is the basis for OLAP analysis and contains information derived from Data Warehouse schema. In the paper, an illustrative case study together with some examples of analysis based on this kind of annotation are provided.