Information systems face significant challenges in today’s constantly evolving digital environments, including dynamic data, heterogeneous sources, and analytical complexities, which directly impact decision-making processes and organizational competitiveness. Data alignment, the process of aligning different sources using their schema and instances, has become a vital solution for ensuring data consistency and enabling effective data exploration. However, existing methods often rely on static approaches which lack adaptability to dynamic data environments and require full recomputation with every change. This study provides an extended evaluation of our previously proposed incremental alignment approach, IDAGEmb, which leverages dynamic graph embedding techniques to refine alignments progressively. Unlike traditional static methods, IDAGEmb adapts to changes in real time, efficiently handling schema modifications and evolving data instances. Our evaluation highlights significant improvements in managing heterogeneous data, optimizing resource usage, and maintaining alignment accuracy in dynamic environments. By integrating incremental graph embeddings, this approach offers a solution for dynamic data environments, providing organizations with consistent and actionable insights. This work builds upon our earlier results, offering a new perspective on data alignment for evolving datasets and emphasizing the effectiveness of dynamic embedding techniques.

Clustering algorithms play a pivotal role in data mining, offering powerful tools for uncovering hidden patterns and structures within datasets. These algorithms aim to divide data points into coherent groups based on similarities or dissimilarities, making it easier to explore and understand complex data. Clustering algorithms typically rely on similarity measures to assess the likeness between data points. Consequently, selecting a suitable similarity measure is crucial for achieving satisfactory clustering outcomes. However, this decision can pose significant challenges, especially for non-experts, given the plethora of similarity measures available in the literature and their performance which is closely linked to the specific dataset, clustering algorithm, and cluster validity index employed. This difficulty is even more important when considering mixed data clustering. Mixed data refers to heterogeneous data characterized by both numerical and categorical attributes. In such a context, the same similarity measure cannot be used for both types of attributes due to their different nature. Commonly, two similarity measures are combined, one for numerical attributes and one for categorical attributes. This adds a layer of complexity to the problem since it requires the selection of two similarity measures instead of just one. This paper introduces SIMREC, a similarity measure recommendation system for mixed data clustering. The system uses meta-learning to mine the relationship between dataset characteristics and similarity measures performances for different mixed data clustering algorithms and cluster validity indices. Therefore, given a mixed dataset, a mixed data clustering algorithm, and a cluster validity index, the system can recommend suitable pairs of numerical and categorical similarity measures based on the characteristics of the dataset. We implemented the proposed system using 130 pairs of similarity measures (10 numerical and 13 categorical), 4 commonly used mixed data clustering algorithms (K-Prototypes, LSH-K-Prototypes, K-Medoids, and Hierarchical Clustering), and three cluster validity indices (Silhouette, Clustering Accuracy, and Adjusted Rand Index). Our experiments on 185 publicly available mixed datasets show that the pairs of similarity measures recommended by SIMREC outperform the baseline pairs, including classically used pairs of similarity measures in the literature.

The legal industry is characterized by the presence of dense and complex documents, which necessitate automatic processing methods to manage and analyse large volumes of data. Traditional methods for extracting legal information depend heavily on substantial quantities of annotated data during the training phase. However, a question arises on how to extract information effectively in contexts that do not favour the utilization of annotated data. This study investigates the application of Large Language Models (LLMs) as a transformative solution for the extraction of legal terms, presenting a novel approach to overcome the constraints associated with the need for extensive annotated datasets. Our research delved into methods such as prompt-engineering and fine-tuning to enhance their performance. We evaluated and compared, to a rule-based and BERT systems, the performance of four LLMs: GPT-4, Miqu-1-70b, Mixtral-8x7b, and Mistral-7b, within the scope of limited annotated data availability. We implemented and assessed our methodologies using Luxembourg's traffic regulations as a case study. Our findings underscore the capacity of LLMs to successfully deal with legal terms extraction, emphasizing the benefits of one-shot and zero-shot learning capabilities in reducing reliance on annotated data by reaching 0.690 F1 Score. Moreover, our study sheds light on the optimal practices for employing LLMs in the processing of legal information, offering insights into the challenges and limitations, including issues related to terms boundary extraction.

This overview traced the evolution of data management, transitioning from traditional ETL processes to addressing contemporary challenges in Big Data, with a particular emphasis on data ingestion and schema matching. It explored the classification of data ingestion into batch, real-time, and hybrid processing, underscoring the challenges associated with data quality and heterogeneity. Central to the discussion was the role of schema mapping in data alignment, proving indispensable for linking diverse data sources. Recent advancements, notably the adoption of machine learning techniques, were significantly reshaping the landscape. The paper also addressed current challenges, including the integration of new technologies and the necessity for effective schema matching solutions, highlighting the continuously evolving nature of schema matching in the context of Big Data.

In the evolving digital environments, information systems are faced with a myriad of challenges such as data heterogeneity, the dynamic nature of data and integration complexities. These challenges impact on decision-making and data integration processes. We define data alignment as the process of aligning columns from different tabular sources using their schema and instances. Data alignment is emerging as an essential solution, ensuring data consistency between different sources and enabling effective integration and decision-making. However, existing solutions fail to take into account the dynamic nature of data in an incremental way. This study presents an incremental methodology that uses dynamic graph embedding techniques to progressively refine data alignments. Although the use of graph embedding techniques for data alignment is well established, their integration into incremental processing approaches remains less explored. This research attempts to fill this gap by evaluating the potential of incremental graph embedding techniques for data alignment. The adoption of this incremental technique has significantly improved the management of heterogeneous data in dynamic environments, while optimizing resource usage. Likewise, this study brings a new perspective to the field of data alignment at it aims to highlight the usefulness of dynamic embedding techniques for the exploration of dynamic datasets.

The legal industry is characterized by the presence of large volumes and complex documents. Given the continuous evolution of these documents, there is a growing interest in automating the processing of legal texts to streamline compliance. One key step of this process is the extraction of legal entities. State-of-the-art methods for legal entity extraction, including rule-based systems, Bi-LSTM, and BERT, require substantial annotated data to be effective, a task that is time-intensive for domain experts. With the rise of Large Language Models (LLMs), research has increasingly focused on leveraging their capabilities and exploring zero-shot approaches. In this paper, we present a hybrid system that distils GPT-4 knowledge through rule-based methods into a CamemBERT model. This approach not only reduces the need for expert involvement compared to the standard CamemBERT system but also outperforms the GPT-4-only system, enhancing the F1 score for legal entities by 9–24

In dynamic information systems, data alignment addresses challenges like data heterogeneity, integration, and interoperability by connecting diverse datasets. To ensure the stability and effectiveness of these alignments over time, an incremental process may be required, allowing the alignments to be updated as the data evolves. While embedding-based methods are valuable for handling incremental data in the graph learning field, they are underexplored in data alignment. However, before implementing such an approach, it is essential to verify the stability of the embeddings in order to guarantee their reliability and temporal consistency. So, we study the most promising model (i.e. Node2Vec) that exhibits favourable stability in embeddings, particularly with respect to the stability of node embeddings. Despite potential variability in pairwise similarities, the idea of an incremental approach remains reliable, especially with a fixed model. Implementing such an approach can efficiently manage data dynamics in information systems with reduced resource needs. By applying this incremental process to data alignment, it will be possible to efficiently manage heterogeneous data in dynamic information system environments, while minimising resource requirements.

In many domains, we face heterogeneous data with both numeric and categorical attributes. Clustering such data is challenging because the notion of similarity is not well defined due to the multiple data types. Existing clustering algorithms for these data are mainly based on two strategies: the homogenization one where all attributes are converted to a single type and the mixed one where similarity measures for the different data types are combined to define a similarity measure for heterogeneous data. We propose a framework in which we evaluate and compare several clustering algorithms using these two strategies on many real-world data sets. Then, motivated by the importance of similarity in clustering and the diversity of similarity measures for each data type, we proposed as a second study, to evaluate how their choice affects the performance of clustering algorithms using the mixed strategy. Our results suggest that the mixed strategy is preferable to the homogenization one since it uses adapted similarity measures for the different data types. Furthermore, the choice of similarity measures is very important for most of used mixed methods and an optimal choice may lead to great improvements compared to classically used similarity measures.

The field of explainability in machine learning has become very prolific and numerous explanation methods have emerged during the last decade. Local explanations are of major interest because they are intelligible and claim to be locally faithful to the machine learning model. However, there is no guarantee that these explanations can apply to new similar observations. This raises questions about the usefulness of acquired knowledge through such explanations. In this paper, we propose a new concept, which we call the validity domain of explanations. This consists in searching for a region around a given local explanation allowing users to determine the extent to which the explanation generalizes. We experimentally show that our approach is more precise regarding the machine learning model and has a better coverage of the dataset than Anchors, a known competitor that paved the way for this issue.

Compliance with legal documents related to industrial maintenance is the company's obligation to oversee, maintain, and repair its equipments. As legal documents endlessly evolve, companies are in favour of automatically processing these texts to facilitate the analysis and compliance. The automatic process involves first, in this pipeline, the extraction of legal entities. However, state-of-the-art, like BERT approaches, have so far required a large amount of data to be effective. Creating this training dataset however is a time-consuming task requiring input from domain experts. In this paper, we bootstrap the legal entity extraction by levering Large Language Models and a semantic model in order to reduce the involvement of the domain experts. We develop the industrial perspective by detailing the technical implementation choices. Consequently, we present our roadmap for an end-to-end pipeline designed expressly for the extraction of legal rules while limiting the involvement of experts.

Data-driven modelling has recently gained interest in the scientific computing community with the purpose of emulating complex large scale systems. Surrogate modelling based on autoencoders (AEs) is widely employed across several engineering fields to model the time-history response of nonlinear high-dimensional dynamical systems from a set of design parameters. In this direction, this paper introduces an efficient deep learning scheme consisting of a two-steps autoencoding framework in conjunction with Neural Ordinary Differential Equations (NODEs), a novel approach for modelling time-continuous dynamics. The proposition aims at alleviating the drawbacks of similar methodologies employed for the same task, namely Parametrized NODE (PNODE) and the two-steps AE-based surrogate models, to provide a more powerful predictive tool. The effectiveness of the conceived methodology has been assessed by considering the task of emulating the spatiotemporal dynamics described by the 1D viscous Burgers’ equation. The outcomes of our empirical analysis demonstrate that our approach outperforms the alternative state-of-the-art models in terms of predictive capability.

A large number of XAI (eXplainable Artificial Intelligence) solutions have been proposed in recent years. Recently, thanks to new XAI evaluation metrics, it has become possible to compare these XAI solutions. However, selecting the most relevant XAI solution among all this diversity is still a tedious task, especially if a user has specific needs and constraints. In this paper, we propose AutoXAI, a framework that recommends the best XAI solution and its hyperparameters according to specified XAI evaluation metrics while considering the user's context (dataset, machine learning model, XAI needs and constraints). It adapts approaches from context-aware recommender systems on one side and strategies of optimization and evaluation from AutoML (Automated Machine Learning) on the other. Through two use cases, we show that AutoXAI recommends XAI solutions adapted to the user's needs with the best hyperparameters matching the user's constraints.

Social media are becoming the preferred channel to report and discuss events happening around the world. The data from these channels can be used to detect ongoing events in real-time. A typical approach is to use event detection methods, usually consisting of a clustering phase, in which similar documents are grouped together, and then an analysis of the clusters to decide whether they deal with real-world events. To cluster together similar documents, content representation models are critical. In this paper, we individually compare the performances of different social media documents content representation models used during the clustering phase, exploiting lexical, semantic and social media specific features, like tags and URLs. To the best of our knowledge, thesemodels are usually individually exploited in this context. We investigate their complementarity and propose to combine them.

It is well known that the construction of traditional reservoir simulation models can be very time and resources consuming. Particularly in the case of mature fields with long history and large number of wells where such models can be extremely difficult and long to history match. In this case data driven models can represent a cost-effective alternative, or they can provide complementary analysis to classical reservoir modelling. Due to data scarcity full machine learning approaches are also usually doomed to fail. In this work we develop a new Physics-Constrained Deep Learning approach that combined neural networks with a reduced physics approach: Capacitance Resistive Model (CRM). CRM are data-driven methods that are based on a simple material balance approximation, that can provide very useful reservoir insight. CRM can be used to analyze the underlying connections between producer wells and injector wells that can then be used to better allocate water injection. Such analysis can usually require very long tracer tests or very expensive 4D seismic acquisition and interpretation. CRM can provide directly these wells connection information using only available production and pressure data. The problem with CRM approaches, based on classical optimizers, is that they often detect spurious correlations and can be not very robust and reliable. Our physics-constrained deep learning approach called Deep-CRM performs production data regularization via the neural network approximation that helps to provide a better CRM parameter identification also with the use of robust gradient descent optimization methods developed and widely used by the large deep learning community. We show first on a synthetic and then in real reservoir case that Deep-CRM was able to identify most of the injector-producer connections with higher accuracy with respect to traditional CRM. Deep-CRM produced also better liquid production forecasts on the performed blind tests.