International audience

With the extensive use of Machine Learning (ML) in the biomedical field, there was an increasing need for Explainable Artificial Intelligence (XAI) to improve transparency and reveal complex hidden relationships between variables for medical practitioners, while meeting regulatory requirements. Feature Selection (FS) is widely used as a part of a biomedical ML pipeline to significantly reduce the number of variables while preserving as much information as possible. However, the choice of FS methods affects the entire pipeline including the final prediction explanations, whereas very few works investigate the relationship between FS and model explanations. Through a systematic workflow performed on 145 datasets and an illustration on medical data, the present work demonstrated the promising complementarity of two metrics based on explanations (using ranking and influence changes) in addition to accuracy and retention rate to select the most appropriate FS/ML models. Measuring how much explanations differ with/without FS are particularly promising for FS methods recommendation. While reliefF generally performs the best on average, the optimal choice may vary for each dataset. Positioning FS methods in a tridimensional space, integrating explanations-based metrics, accuracy and retention rate, would allow the user to choose the priorities to be given on each of the dimensions. In biomedical applications, where each medical condition may have its own preferences, this framework will make it possible to offer the healthcare professional the appropriate FS technique, to select the variables that have an important explainable impact, even if this comes at the expense of a limited drop of accuracy.

Data lakes are supposed to enable analysts to perform more efficient and efficacious data analysis by crossing multiple existing data sources, processes and analyses. However, it is impossible to achieve that when a data lake does not have a metadata governance system that progressively capitalizes on all the performed analysis experiments. The objective of this paper is to have an easily accessible, reusable data lake that capitalizes on all user experiences. To meet this need, we propose an analysis-oriented metadata model for data lakes. This model includes the descriptive information of datasets and their attributes, as well as all metadata related to the machine learning analyzes performed on these datasets. To illustrate our metadata solution, we implemented an application of data lake metadata management. This application allows users to find and use existing data, processes and analyses by searching relevant metadata stored in a NoSQL data store within the data lake. To demonstrate how to easily discover metadata with the application, we present two use cases, with real data, including datasets similarity detection and machine learning guidance.

The issue of democracy in society is at the heart of our current concerns. Organizations and their information systems are also concerned by this issue. Democracy in organization requires a debate about norms, values and language encapsulated in the information system. Participatory design approaches address this issue by proposing a democratic empowerment for users during design phase of projects. To go further, we propose a structured method to integrate democracy into information system. This method named DEMOS for DEsign Method for demOcratic information System is described and then illustrated by a real experiment provided by a “lifelong training” service at the University.

As the amount of data grows very fast inside and outside of an enterprise, it is getting important to analyze both of them for getting total business intelligence. While online analytical processing (OLAP) techniques have been proven very useful for analyzing structured data, they face challenges in handling unstructured data. To this end, new multidimensional models have been proposed for OLAP purposes. Nevertheless, there is no proposal allowing managing both document structures and the semantics of the textual content. In our previous work, we proposed to integrate the entire document within a Diamond multi-dimensional model. In this paper, based on our proposed model, we provide new OLAP operators that take into account the specificities of this model.

Organizational memory is a space where various information circulating in a company are capitalized. From the users' point of view, an organizational memory, which can be seen as an information system component, is very important since it stores the “shared knowledge” of the organization. But, at the same time, the cost of this knowledge is relatively high since users' participation, i.e. to integrate/maintain... the memory is important. The aim of our work is to model an organizational memory through a heterogeneous network on which is based an automatic information integration process to assist users in this task while limiting their effort. We developed a prototype and evaluated through an experiment its ability to integrate new information into an organizational memory based on the proposed model.

On-Line Analytical Processing (OLAP) has generated methodologies for the analysis of structured data. However, they are not appropriate to handle document content analysis. Because of the fast growing of this type of data, there is a need for new approaches abling to manage textual content of data. Generally, these data exist in XML format. In this context, we propose an approach of construction of our Diamond multidimensional model, which includes semantic dimension to better consider the semantics of textual data In addition, we propose new aggregation operators for textual data in OLAP environment.

International audience

The quantity of digital documents available is still growing. The various contexts of use of such documents need several kinds of descriptions of their contents and structures. Thus a same document can be described according to several concurrent structures. Designing models and tools to exploit these various kinds of structures simultaneously presents a real challenge. In this way we have built document repositories to achieve this aim. Indeed, we proposed fragmentation techniques to manage the various issues related to the management of multistructured documents (representation, storage, reconstruction, and management of concurrent structures). This paper is dedicated to the presentation of the formal model. We propose to describe with precision and concision the various concepts related to the multi-structured documents as well as rules related to the organization of these documents.

This article proposes an approach of construction an ontological intentions based on the work study suggested in the literature. Many methods were proposed in the literature. We are interested to the methods applying to the texts, and more particularly, with the methods to learn ontologies starting from textual corpus. This paper presents a construction of an intentions ontology. We state some fundamental principles to respect to build an ontology. We will build the ontology of specific domain which comprises an abstraction levels to knowing a linguistics level, and understanding the effective concepts of the domain and the relations which links them, will be useful for the knowledge extraction between the various semantic contexts, knowing a semantic context is define as being a subset of concepts and relations which link them. The Ontology of Verb Concepts combines linguistic and psycholinguistic classification. The information for verbs encodes typical associations with actions and events. We propose a method by using domain ontology and syntactic-semantic analysis. This article presents experimentation on scientific publications article in the domain of computer science.