Currently, the need for business documents analysis, particularly invoices, is playing a vital role in companies, especially in large ones. These documents have the particularity of being visually rich, with low text quantity and many different layouts. As such, processing them with traditional techniques remains inefficient. Hence, one of the key challenge is to exploit visual patterns between entities of interest. After an overview of the state-of-the-art in this domain, we propose a graph-based model that recognizes specific text in invoices. First, an Encoder module creates a multimodal embedding for each text sequence based on textual, visual, and spatial information. This representation is then passed through a multi-layer graph attention network, before being subjected to a simple classification task. Some experimental results were conducted in order to improve the performance of the proposed approach.

Regulations such as GDPR and CCPA are making the management of personal information an increasingly sensitive issue for businesses and organizations. We consider that data protection mechanisms must intervene upstream as recommended by the principles of privacy by design. Our view is that database management systems should be equipped with anonymization mechanisms. By relying on a partitioned data storage scheme, support for data update operations and query rewriting, our approach offers a trade-off between data privacy and utility. Some advantages of our approach are a low overhead of storing partitioned data, the performance of CRUD operations and the efficiency of a batch delivery approach to data updates. We evaluate our prototype on an open source RDF store and on synthetic datasets.

International audience

As the usage of RDF-based Knowledge Graphs is going mainstream, it becomes necessary for organizations and companies to consider the privacy preservation of the data they are managing and possibly sharing. This is generally performed by anonymization techniques such as triple suppression and generalization. Nevertheless, these techniques have the drawback of reducing the utility of the released datasets. This paper presents semantic anatomization, a novel anonymization technique, that retains all quasi-identifier and sensitive values in the RDF graph. Due to an aggregating mechanism and the exploitation of the semantics contained in ontologies, this technique preserves data correlation and supports high quality analysis from anonymized graphs. We demonstrate the potential of semantic anatomization on large graphs generated from our own extension of the well-established Lehigh university benchmark.

The overflowing of textual data on the web needs an efficient tool that is able to manage and process data. In this context, automatic text summarization has shown a great importance in several application areas. It aims to create a coherent and fluent short version of a document while preserving of the main information. This method allows for a reduction in reading time by condensing relevant information from a large collection of documents. Several automatic text summarization approaches have been proposed in order to entail shorten parts of the document. These methods have good results, but they still need improvements related to the reliability of sentences extraction, redundancy, semantic relationships between sentences, etc. This paper introduces a new hybrid architecture, combining a 2-layer recurrent neural network (RNN) extractive model and a sequence-to-sequence attentional abstractive model. This method uses the advantages of both extractive and abstractive approaches. A given text is first fed into the extractive model to obtain its relevant content, then generalized using the abstractive model, resulting in the text's summary. First experimental results on real-world data show that the proposed model can achieve competitive results for extractive, abstractive and hybrid models.

As edge computing is becoming a new platform for rich applications and services, it becomes more and more important to design adapted data management systems for this environment. In this paper, we present a prototype corresponding to a compact, in-memory RDF store that can answer SPARQL queries requiring reasoning services without necessitating any decompression. This demonstration highlights a design based on succinct data structures, shows some implementation details and provides encouraging performance measures over a set of real-world and synthetic data and query sets.

Text detection and recognition have witnessed drastic improvements in the field of computer vision. This end-toend model comprising of the detection and recognition models scales to provide higher accuracy. The most important phase in this end-to-end approach is the detection phase, as it plays an important role to identify the text. To address this issue, different approaches have been proposed. However, most of the methods produce lower efficiency to detect and recognize real world text. In this paper, we propose a new approach to investigate the challenges that the existing models possess and improve the efficiency of the detection and in turn increases the accuracy of text recognition. The proposed method outperforms the state-ofthe-art approaches due to the use of deblurring and sharpening to reduce noise in the pre-processing stage, followed by the cascade region proposal network model to improve the detection of real world text using non max suppression. Experimentations on real word datasets highlight the effectiveness of our method.

Worldwide, human activities have a major impact on energy production and consumption. Urban areas, where the majority of the world population lives, are confronted with many environmental problems especially in emerging countries where a potential ecological transition is shadowed by frenetic economic development. At the same time, the deployment of intelligent infrastructures (Smart Grids), new technologies or paradigms (ubiquitous computing, Big Data) impact behaviors and practices of inhabitants of these areas. The ability to aggregate and model these digital traces in multiple dimensions could allow people to better understand their daily activities and promote more sustainable behaviors and practices by reducing the footprints related to every human collective activity. This paper aims to explore how to facilitate the decision-making process for inhabitants of these intelligent urban areas about their sustainable practices and lifestyles based on massive heterogeneous data in order to optimize the daily production and consumption of energy and meet the challenges of energy access and ecological transition.

Sustainable behaviors, social interaction, mobility and information aggregation inside digital business environment need to converge to reach the next step of collaboration to enhance collaboration and innovation. The following article is based on the “Assemblage” concept seen as a framework to formalize new user interfaces and applications. The area of research is the Energy Social Business Environment and Smart Cities, especially the Energy Smart Grids, which are considered as functional and technical foundations of the revolution of the Energy Sector of tomorrow. The main goal is to offer new central attention and decision-making tools to end-users to help them to better use Energy.