Search

International audience; K-anonymity is the most widely used approach to privacy preserving microdata which is mainly based on generalization. Although generalization-based k-anonymity approaches can achieve the privacy protection objective, they suffer from information loss. Clustering-based approaches have been successfully adapted for k-anonymization as they enhance the data quality, however, the computational complexity of finding an optimal solution has shown as NP-hard. Nature-inspired optimization algorithms are effective in finding solutions to complex problems. We propose, in this paper, a novel algorithm based on a simple nature-inspired metaheuristic called Black Hole Algorithm (BHA), to address such limitations. Experiments on real data set show that data utility has been improved by our approach compared to k-anonymity, BHA-based k-anonymity and clustering-based k-anonymity approaches.

A major drawback of nonlinear dimensionality reduction (DR) techniques is their inability to preserve some authentic information from the source domain, leading to projections that are often hard to interpret when it comes to observing anything other than the topological structure of the data. In this paper, we propose a nonlinear DR approach enforcing projection constraints resulting from an a priori knowledge about the structure of the data in multivariate periodical time series. We then propose several ways of exploiting this constrained projection to extract user-relevant information, such as the nominal behavior of a periodical dynamical system or the deviant behaviors which may occur at different time scales. The techniques are demonstrated on both a synthetic dataset composed of simulated multivariate data exhibiting a periodical behavior, and a real dataset corresponding to six months of sensor data acquisitions and recordings inside experimental buildings. 1 We would like to thank the Institut National de l'Energie Solaire (INES) and the CEA, LITEN, Laboratoire Energetique du Bâtiment for providing the data.

International audience; Manufacturing industry needs access to the data in order to realise its activities but also to generate new value-added knowledge. Nevertheless, it is confronted with a large and growing volume of heterogeneous data which limits its ability to exploit them optimally. Moreover, the data are distributed within different heterogeneous information systems, which limits the relationship exploration under the information retrieval process. Usually, the challenge is addressed by trying to manage and normalize the data structure in order to faster searching and exploiting them in a manufacturing context. For their part, the authors present i-Dataquest, an information retrieval system supported by (i) a graph-oriented model built from the structured and unstructured data of the company and (ii) a query system answering ‘what’ and ‘about what’ and (iii) generating three different results: a list of items, a list of property values and a list of sentences. The i-Dataquest prototype is built using Neo4J for the graph system generation, ConceptNet for lexical resource management and StandfordNLP for natural language processing. An evaluation of the prototype’s performance is conducted through a data set representing a drone manufacturer. The results show that the transformation of specific content such as tables in the graph and the semantic expansion of queries significantly improves the recall and precision measures. The results also suggest improving filtering less relevant results by considering particularly queries looking for a specific value.

This brief addresses understandability of modern machine learning networks with respect to the statistical properties of their convolution layers. It proposes a set of tools for categorizing a convolution layer in terms of kernel property (meanlet, differencelet, or distrotlet) or kernel sequence property (frame spectra and intralayer correlation matrix). These tools are expected to be relevant for determining the generalization capabilities of a convolutional neural network. In particular, this brief highlights that the less frequency penalizing network among AlexNet, GoogleNet, RESNET101, and VGG19 is the more relevant one in terms of solutions for low-level ice-sheet feature enhancement.

In today’s highly connected cyber-physical environments, users are becoming more and more concerned about their privacy and ask for more involvement in the control of their data. However, achieving effective involvement of users requires improving their privacy decision-making. This can be achieved by: (i) raising their awareness regarding the direct and indirect privacy risks they accept to take when sharing data with consumers; (ii) helping them in optimizing their privacy protection decisions to meet their privacy requirements while maximizing data utility. In this article, we address the second goal by proposing a user-centric multi-objective approach for context-aware privacy management in connected environments, denoted δ- Risk . Our approach features a new privacy risk quantification model to dynamically calculate and select the best protection strategies for the user based on her preferences and contexts. Computed strategies are optimal in that they seek to closely satisfy user requirements and preferences while maximizing data utility and minimizing the cost of protection. We implemented our proposed approach and evaluated its performance and effectiveness in various scenarios. The results show that δ- Risk delivers scalability and low-complexity in time and space. Besides, it handles privacy reasoning in real-time, making it able to support the user in various contexts, including ephemeral ones. It also provides the user with at least one best strategy per context.

International audience; This paper presents parallel solutions (developed based on two state-of-the-art algorithms iSAX and sketch) for evaluating kNN (k nearest neighbor) queries on large databases of time series, compares them based on various measures of quality and time performance, offers a tool that uses the characteristics of application data to determine which algorithm to choose for that application and how to set the parameters for that algorithm. Specifically, our experiments show that: (i) iSAX and its derivatives perform best in both time and quality when the time series can be characterized by a few low frequency Fourier Coefficients , a regime where the iSAX pruning approach works well. (ii) iSAX performs significantly less well when high frequency Fourier Coefficients have much of the energy of the time series. (iii) A random projection approach based on sketches by contrast is more or less independent of the frequency power spectrum. The experiments show the close relationship between pruning ratio and time for exact iSAX as well as between pruning ratio and the quality of approximate iSAX. Our toolkit analyzes typical time series of an application (i) to determine optimal segment sizes for iSAX and (ii) when to use Parallel Sketches instead of iSAX. Our algorithms have been implemented using Spark, evaluated over a cluster of nodes, and have been applied to both real and synthetic data. The results apply to any databases of numerical sequences, whether or not they relate to time.

Supervised deep learning has achieved remarkable success in various applications. Successful machine learning application however depends on the availability of sufficiently large amount of data. In the absence of data from the target domain, representative data collection from multiple sources is often needed. However, a model trained on existing multi-source data might generalize poorly on the unseen target domain. This problem is referred to as domain shift. In this paper, we explore the suitability of multi-source training data selection to tackle the domain shift challenge in the context of domain generalization. We also propose a microservice-oriented methodology for supporting this solution. We perform our experimental study on the use case of building energy consumption prediction. Experimental results suggest that minimal building description is capable of improving cross-building generalization performances when used to select energy consumption data., Service Oriented Computing and Applications, Springer, 2020

Directed acyclic graphs (DAGs) are used in many domains ranging from computer science to bioinformatics, including industry and geoscience. They enable to model complex evolutions where spatial objects (e.g., soil erosion) may move, (dis)appear, merge or split. We study a new graph-based representation, called attributed DAG (a-DAG). It enables to capture interactions between objects as well as information on objects (e.g., characteristics or events). In this paper, we focus on pattern mining in such data. Our patterns, called weighted paths, offer a good trade-off between expressiveness and complexity. Frequency and compactness constraints are used to filter out uninteresting patterns. These constraints lead to an exact condensed representation (without loss of information) in the single-graph setting. A depth-first search strategy and an optimized data structure are proposed to achieve the efficiency of weighted path discovery. It does a progressive extension of patterns based on database projections. Relevance, scalability and genericity are illustrated by means of qualitative and quantitative results when mining various real and synthetic datasets. In particular, we show how such an approach can be used to monitor soil erosion using remote sensing and geographical information system (GIS) data.

The necessity to improve security in a multi-cloud environment has become very urgent in recent years. Although in this topic, many methods using the message authentication code had been realized but, the results of these methods are unsatisfactory and heavy to apply, which, is why the security problem remains unresolved in this environment. This article proposes a new model that provides authentication and data integrity in a distributed and interoperable environment. For that in this paper, the authors first analyze some security models used in a large and distributed environment, and then, we introduce a new model to solve security issues in this environment. Our approach consists of three steps, the first step, was to propose a private virtual network to secure the data in transit. Secondly, we used an authentication method based on data encryption, to protect the identity of the user and his data, and finally, we realize an algorithm to know the integrity of data distributed on the various clouds of the system. The model achieves both identity authentication and the ability to inter-operate between processes running on different cloud’s provider. A data integrity algorithm will be demonstrated. The results of this proposed model can efficiently and safely construct a reliable and stable system in the cross-cloud environment.

In this paper, our goal is to investigate the behavior of reassignment operators used in synchrosqueezing transforms to reassign the time-frequency representation of multicomponent signals made of the superposition of amplitude and frequency modulated modes. Indeed, while these operators are particularly efficient on specific types of modes, their quality worsens drastically when the modes depart from the ideal case they are designed for. In particular, when these interfere in the time-frequency plane or when some noise is present, we show the limits to the use of these reassignment operators for the estimation of the instantaneous frequency of the modes by studying their behavior in the vicinity of spectrogram ridges, and then propose a novel approach to circumvent these limitations.

International audience; The paper presents the cartographic processing of the Landsat TM image by the two unsupervised classification methods of SAGA GIS: ISODATA and K-means clustering. The approaches were tested and compared for land cover type mapping. Vegetation areas were detected and separated from other land cover types in the study area of southwestern Iceland. The number of clusters was set to ten classes. The processing of the satellite image by SAGA GIS was achieved using Imagery Classification tools in the Geoprocessing menu of SAGA GIS. Unsupervised classification performed effectively in the unlabeled pixels for the land cover types using machine learning in GIS. Following an iterative approach of clustering, the pixels were grouped in each step of the algorithm and the clusters were reassigned as centroids. The paper contributes to the technical development of the application of machine learning in cartography by demonstrating the effectiveness of SAGA GIS in remote sensing data processing applied for vegetation and environmental mapping.

[EN] We define here indirectly named entities, as a term to denote multiword expressions referring to known named entities by means of periphrasis. While named entity recognition is a classical task in natural language processing, little attention has been paid to indirectly named entities and their treatment. In this paper, we try to address this gap, describing issues related to the detection and understanding of indirectly named entities in texts. We introduce a proof of concept for retrieving both lexicalised and non-lexicalised indirectly named entities in French texts. We also show example cases where this proof of concept is applied, and discuss future perspectives. We have initiated the creation of a first lexicon of 712 indirectly named entity entries that is available for future research., This research has been funded by the FEDER (Fonds européen de développement régional) and selected by the French-Swiss programme Interreg V. We would like to thank Claire Wuillemin for her preliminary work in the DecRIPT project about the State-of-the-Art in NER and SER in 2020. We would also like to thank for their advice Gilles Falquet, Luka Nerima, Eric Wehrli and Jean-Philippe Goldman at the University of Geneva.

International audience; This paper focuses on similarity caching systems, in which a user request for an object o that is not in the cache can be (partially) satisfied by a similar stored object o , at the cost of a loss of user utility. Similarity caching systems can be effectively employed in several application areas, like multimedia retrieval, recommender systems, genome study, and machine learning training/serving. However, despite their relevance, the behavior of such systems is far from being well understood. In this paper, we provide a first comprehensive analysis of similarity caching in the offline, adversarial, and stochastic settings. We show that similarity caching raises significant new challenges, for which we propose the first dynamic policies with some optimality guarantees. We evaluate the performance of our schemes under both synthetic and real request traces.

A particularly challenging problem for data anonymization is dealing with transactional data. Most anonymization methods assume homogeneous, independent and identically distributed (i.i.d.) data; “flattening” transactional data to satisfy this model results in wide, sparse data that does not anonymize well with traditional techniques. While there have been some approaches for generalization-based anonymization, bucketization techniques (e.g., anatomy) pose new challenges. In particular, bucketization provides the opportunity to learn correlations between data items, but also a risk of identifying individuals because of dependencies inferred from such correlations. We present a method that balances these issues, retaining the ability to discover correlations in the data, while hiding dependencies that would enable correlations to be used to link specific values to individuals. We introduce a correlation anonymization constraint that ensures correlations do not allow data to be linked to a specific individual, and an elastic safe grouping algorithm that meets this constraint while preserving data correlations. We evaluate the utility loss on a transactional rental dataset.

Many efforts have been deployed by the IR community to extend free-text query processing toward semi-structured XML search. Most methods rely on the concept of Lowest Comment Ancestor (LCA) between two or multiple structural nodes to identify the most specific XML elements containing query keywords posted by the user. Yet, few of the existing approaches consider XML semantics, and the methods that process semantics generally rely on computationally expensive word sense disambiguation (WSD) techniques, or apply semantic analysis in one stage only: performing query relaxation/refinement over the bag of words retrieval model, to reduce processing time. In this paper, we describe the building blocks of a new approach for XML keyword search aiming to solve the limitations mentioned above. Our solution first transforms the XML document collection (offline) and the keyword query (on-the-fly) into meaningful semantic representations using context-based and global disambiguation methods, specially designed to allow almost linear computation efficiency. Consequently, the semantically augmented XML data tree is processed for structural node clustering, based on semantic query concepts (i.e., key-concepts), in order to identify and rank candidate answer sub-trees containing related occurrences of query key-concepts. Preliminary experiments highlight the quality and potential of our approach.

International audience; Ontology alignment plays a key role in the management of heterogeneous data sources and metadata. In this context, various ontology alignment techniques have been proposed to discover correspondences between the entities of different ontologies. This paper proposes a new ontology alignment approach based on a set of rules exploiting the embedding space and measuring clusters of labels to discover the relationship between entities. We tested our system on the OAEI conference complex alignment benchmark track and then applied it to aligning ontologies in a real-world case study. The experimental results show that the combination of word embedding and a measure of dispersion of the clusters of labels, which we call the radius measure, makes it possible to determine, with good accuracy, not only equivalence relations, but also hierarchical relations between entities.

International audience; Crowdsourcing is a way to solve problems that need human contribution. Crowdsourcing platforms distribute replicated tasks to workers, pay them for their contribution, and aggregate answers to produce a reliable conclusion. A fundamental problem is to infer a consensual answer from the set of returned results. Another problem is to obtain this answer at a reasonable cost: unlimited budget allows hiring experts or large pools of workers for each task but a limited budget forces to use resources at best. Last, crowdsourcing platforms have to detect and ban malevolent users (also known as "spammers") to achieve good accuracy of their answers. This paper considers crowdsourcing of simple Boolean tasks. We first define a probabilistic inference technique, that considers difficulty of tasks and expertise of workers when aggregating answers. We then propose CrowdInc, a greedy algorithm that reduces the cost needed to reach a consensual answer. CrowdInc distributes resources dynamically to tasks according to their difficulty. The algorithm solves batches of simple tasks in rounds that estimate workers expertize, tasks difficulty, and synthesizes a plausible aggregated conclusion and a confidence score using Expectation Maximization. The synthesized values are used to decide whether more workers should be hired to increase confidence in synthesized answers. We show on several benchmarks that CrowdInc achieves good accuracy, reduces costs and we compare its performance to existing solutions. We then use the estimation of CrowdInc to detect spammers and study the impact of spammers on costs and accuracy.

In the Web of Things (WoT) context, an increasing number of stationary and mobile objects provide functions as RESTful services, also called resources, that can be combined with other existing Web resources, to create value-added processes. However, nowadays resource discovery and selection are challenging, due to (1) the growing number of resources providing similar functions, making Quality of Resource (QoR) essential to select appropriate resources, (2) the transient nature of resource availability due to sporadic connectivity, and (3) the location changes of mobile objects in time. In this paper, we first present a location-aware resource discovery that relies on a 3-dimensional indexing schema, which considers object location for resource identification. Then, we present a QoR-driven resource selection approach that uses a Selection Strategy Adaptor (SSA) to form i-compositions (with i ∈N*) offering different implementation alternatives. The defined SSA allows forming resource compositions while considering QoR constraints and Inputs/Outputs matching of related resources, as well as resource availability and users different needs (e.g., optimal and optimistic compositions obtained using a scoring system). Analyses are made to evaluate our service quality model against existing ones, and experiments are conducted in different environments setups to study the performance of our solution.

Evaluating the use and impact of telemedicine in nursing homes is necessary to promote improvements in the quality of this practice. Even though challenges and opportunities of telemedicine are increasingly becoming well documented for geriatrics (such as improving access to healthcare, patient management, and education while reducing costs), there is still limited knowledge on how to better implement it in an inter-organizational context, especially when considering nursing homes. In this regard, this study aimed first to describe the telemedicine activity of nursing homes when cooperating with a general hospital; and then understand the behavioral differences amongst nursing homes while identifying critical factors when implementing a telemedicine project. We conducted a sequential, explanatory mixed-method study using quantitative then qualitative methods to better understand the results. Three years of teleconsultation data of twenty-six nursing homes (15 rural and 11 urban) conducting teleconsultations with a general hospital (Troyes Hospital, France) were included for the quantitative analysis, and eleven telemedicine project managers for the qualitative analysis. Between April 2018 and April 2021, 590 teleconsultations were conducted: 45% (n = 265) were conducted for general practice, 29% (n = 172) for wound care, 11% (n = 62) for diabetes management, 8% (n = 47) with gerontologist and 6% (n = 38) for dermatology. Rural nursing homes conducted more teleconsultations overall than urban ones (RR: 2.484; 95% CI: 1.083 to 5.518; p = 0.03) and included more teleconsultations for general practice (RR: 16.305; 95% CI: 3.505 to 73.523; p = 0.001). Our qualitative study showed that three critical factors are required for the implementation of a telemedicine project in nursing homes: (1) the motivation to perform teleconsultations (in other words, improving access to care and cooperation between professionals); (2) building a relevant telemedicine medical offer based on patients’ and treating physicians’ needs; and (3) it’s specific organization in terms of time and space. Our study showed different uses of teleconsultations according to the rural or urban localization of nursing homes and that telemedicine projects should be designed to consider this aspect. Triggered by the COVID-19 pandemic, telemedicine projects in nursing homes are increasing, and observing the three critical factors presented above could be necessary to limit the failure of such projects.

International audience; In most real-world applications, it is seldom the case that a result appears independently from an environment. In social networks, users' behavior results from the people they interact with, news in their feed, or trending topics. In natural language, the meaning of phrases emerges from the combination of words. In general medicine, a diagnosis is established on the basis of the interaction of symptoms. Here, we propose the Interacting Mixed Membership Stochastic Block Model (IMMSBM), which investigates the role of interactions between entities (hashtags, words, memes, etc.) and quantifies their importance within the aforementioned corpora. We find that in inference tasks, taking them into account leads to average relative changes with respect to non-interacting models of up to 150% in the probability of an outcome and greatly improves the predictions performances. Furthermore, their role greatly improves the predictive power of the model. Our findings suggest that neglecting interactions when modeling real-world phenomena might lead to incorrect conclusions being drawn.

Also part of the Information Systems and Applications, incl. Internet/Web, and HCI book sub series (LNISA, volume 12925); International audience; The problem of identifying the degree of semantic similarity between two textual statements automatically has grown in importance in recent times. Its impact on various computer-related domains and recent breakthroughs in neural computation has increased the opportunities for better solutions to be developed. This research takes the research efforts a step further by designing and developing a novel neurofuzzy approach for semantic textual similarity that uses neural networks and fuzzy logics. The fundamental notion is to combine the remarkable capabilities of the current neural models for working with text with the possibilities that fuzzy logic provides for aggregating numerical information in a tailored manner. The results of our experiments suggest that this approach is capable of accurately determining semantic textual similarity.

In this paper, we are interested in temporal sequential data analysis using GALACTIC, a new framework based on Formal Concept Analysis (FCA) for calculating a concept lattice from heterogeneous and complex data. Inspired by pattern structure theory, GALACTIC mines data described by predicates and is composed of a system of plugins for an easy integration of new data characteristics and their descriptions. Here we use the GALACTIC library to analyse temporal sequential data, where each item \(x_i\) in the sequence has an associated timestamp \(t_i\): \(s = \langle (t_i, x_i) \rangle _{i\le n}\). We introduce descriptions and strategies dedicated to temporal sequences, and new unsupervised measures. We show on some datasets that the distance constraint subsequence strategy allows to generate good concept lattices.

Interactions between pieces of information (entities) play a substantial role in the way an individual acts on them: adoption of a product, the spread of news, strategy choice, etc. However, the underlying interaction mechanisms are often unknown and have been little explored in the literature. We introduce an efficient method to infer both the entities interaction network and its evolution according to the temporal distance separating interacting entities; together, they form the interaction profile. The interaction profile allows characterizing the mechanisms of the interaction processes. We approach this problem via a convex model based on recent advances in multi-kernel inference. We consider an ordered sequence of exposures to entities (URL, ads, situations) and the actions the user exerts on them (share, click, decision). We study how users exhibit different behaviors according to combinations of exposures they have been exposed to. We show that the effect of a combination of exposures on a user is more than the sum of each exposure’s independent effect–there is an interaction. We reduce this modeling to a non-parametric convex optimization problem that can be solved in parallel. Our method recovers state-of-the-art results on interaction processes on three real-world datasets and outperforms baselines in the inference of the underlying data generation mechanisms. Finally, we show that interaction profiles can be visualized intuitively, easing the interpretation of the model.

International audience; Vessel structures such as retinal vasculature are important features for computer-aided diagnosis. In this paper, a probabilistic tracking method is proposed to detect blood vessels in retinal images. During the tracking process, vessel edge points are detected iteratively using local grey level statistics and vessel's continuity properties. At a given step, a statistic sampling scheme is adopted to select a number of vessel edge points candidates in a local studying area. Local vessel's sectional intensity profiles are estimated by a Gaussian shaped curve. A Bayesian method with the Maximum a posteriori (MAP) probability criterion is then used to identify local vessel's structure and find out the edge points from these candidates. Evaluation is performed on both simulated vascular and real retinal images. Different geometric shapes and noise levels are used for computer simulated images, whereas real retinal images from the REVIEW database are tested. Evaluation performance is done using the Segmentation Matching Factor (SMF) as a quality parameter. Our approach performed better when comparing it with Sun's and Chaudhuri's methods. ROC curves are also plotted, showing effective detection of retinal blood vessels (true positive rate) with less false detection (false positive rate) than Sun's method

International audience; Computing aggregation over sliding windows, i.e., finite subsets of an unbounded stream, is a core operation in streaming analytics. We propose PBA (Parallel Boundary Aggregator), a novel parallel algorithm that groups continuous slices of streaming values into chunks and exploits two buffers, cumulative slice aggregations and left cumulative slice aggregations, to compute sliding window aggregations efficiently. PBA runs in (1) time, performing at most 3 merging operations per slide while consuming () space for windows with partial aggregations. Our empirical experiments demonstrate that PBA can improve throughput up to 4× while reducing latency, compared to state-of-the-art algorithms.

International audience; BackgroundPubMed is one of the most important basic tools to access medical literature. Semantic query expansion using synonyms can improve retrieval efficacy.ObjectiveThe objective was to evaluate the performance of three semantic query expansion strategies.MethodsQueries were built for forty MeSH descriptors using three semantic expansion strategies (MeSH synonyms, UMLS mappings, and mappings created by the CISMeF team), then sent to PubMed. To evaluate expansion performances for each query, the first twenty citations were selected, and their relevance were judged by three independent evaluators based on the title and abstract.ResultsQueries built with the UMLS expansion provided new citations with a slightly higher mean precision (74.19%) than with the CISMeF expansion (70.28%), although the difference was not significant. Inter‐rater agreement was 0.28. Results varied greatly depending on the descriptor selected.DiscussionThe number of citations retrieved by the three strategies and their precision varied greatly according to the descriptor. This heterogeneity could be explained by the quality of the synonyms. Optimal use of these different expansions would be through various combinations of UMLS and CISMeF intersections or unions.ConclusionInformation retrieval tools should propose different semantic expansions depending on the descriptor and the search objectives.

International audience

International audience

International audience; An intensive recent research work investigated the combined use of hand-curated knowledge resources and corpus-driven resources to learn effective text representations. The overall learning process could be run by online revising the learning objective or by offline refining an original learned representation. The differentiated impact of each of the learning approaches on the quality of the learned representations has not been studied so far in the literature. This article focuses on the design of comparable offline vs. online knowledge-enhanced document representation learning models and the comparison of their effectiveness using a set of standard IR and NLP downstream tasks. The results of quantitative and qualitative analyses show that (1) offline vs. online learning approaches have dissimilar result trends regarding the task as well as the dataset distribution counts with regard to domain application; (2) while considering external knowledge resources is undoubtedly beneficial, the way used to express relational constraints could affect semantic inference effectiveness. The findings of this work present opportunities for the design of future representation learning models, but also for providing insights about the evaluation of such models.

Purpose Visual Distinctive Language (VDL)-based iconic tags are structured visual information annotation. They explicate the content and organization of tagged information by graphical and symbolic features in order to improve the vocabulary problems of textual tags. The purpose of this paper is to investigate how these special icons help in tagged-based user information searching. Design/methodology/approach A two-stage experiment was designed and conducted so as to follow and quantify the searching process in specific searching target case and no specific searching target case when using VDL-based iconic tags. Findings The experimental results manifested that VDL-based iconic tags enhanced the role of tag in information searching. They could make user better understand tag clusters, which, in turn, provide global structure of involved topics. Also, VDL-based iconic tags helped user to find out searching target more quickly with higher accuracy by taking advantages of visual representation of tag categories and symbolic signification of tag content. Originality/value This study is one of the first to verify how structured icons work in information searching and how user’s graphical cognition impacts on tag-based information searching process. The research findings are dedicated to the theory of VDL-based iconic tags, as well as to a new visualization method for search user interface design.

Sentence representation at the semantic level is a challenging task for natural language processing and Artificial Intelligence. Despite the advances in word embeddings (i.e. word vector representations), capturing sentence meaning is an open question due to complexities of semantic interactions among words. In this paper, we present an embedding method, which is aimed at learning unsupervised sentence representations from unlabeled text. We propose an unsupervised method that models a sentence as a weighted series of word embeddings. The weights of the series are fitted by using Shannon’s Mutual Information (MI) among words, sentences and the corpus. In fact, the Term Frequency–Inverse Document Frequency transform (TF–IDF) is a reliable estimate of such MI. Our method offers advantages over existing ones: identifiable modules, short-term training, online inference of (unseen) sentence representations, as well as independence from domain, external knowledge and linguistic annotation resources. Results showed that our model, despite its concreteness and low computational cost, was competitive with the state of the art in well-known Semantic Textual Similarity (STS) tasks.

International audience; A number of importantapplications relatedto complexnetwork analysisrequire finding small vertex covers in massive graphs. This paper proposes an effectivestochastic local searchalgorithm called DTS_MVC to fulfill this task. Relying on a fast vertex-based search strategy, DTS_MVC effectively explores the search space by alternating between a thresholding search phase during which the algorithm accepts both improving and non-improving solutions that satisfy a dynamically changing quality threshold, and a conditional improving phase where only improving solutions are accepted. A novel non-parametric operation-prohibiting mechanism is introduced to avoid search cycling.Computational experimentson 86 massive real-world benchmark graphs indicate that DTS_MVC performs remarkably well by discovering 7 improved best known results (new upper bounds). Additional experiments are conducted to shed light on the key ingredients of DTS_MVC.

International audience; The digitization of music, the emergence of online streaming platforms and mobile apps have dramatically changed the ways we consume music. Today, much of the music that we listen to is organized in some form of a playlist, and many users of modern music platforms create playlists for themselves or to share them with others. The manual creation of such playlists can however be demanding, in particular due to the huge amount of possible tracks that are available online. To help users in this task, music platforms like Spotify provide users with interactive tools for playlist creation. These tools usually recommend additional songs to include given a playlist title or some initial tracks. Interestingly, little is known so far about the effects of providing such a recommendation functionality. We therefore conducted a user study involving 270 subjects, where one half of the participants-the treatment group-were provided with automated recommendations when performing a playlist construction task. We then analyzed to what extent such recommendations are adopted by users and how they influence their choices. Our results, among other aspects, show that about two thirds of the treatment group made active use of the recommendations. Further analyses provide additional insights about the underlying reasons why users selected certain recommendations. Finally, our study also reveals that the mere presence of the recommendations impacts the choices of the participants, even in cases when none of the recommendations was actually chosen.

International audience; While graph-based collaborative filtering recommender systems have been introduced several years ago, there are still several shortcomings to deal with, the temporal information being one of the most important. The new link stream paradigm is aiming at extending graphs for correctly modelling the graph dynamics, without losing crucial information. We investigate the impact of such link stream features for recommender systems. We design link stream features, that capture the intrinsic structure and dynamics of the data. We show that such features encode a fine-grained and subtle description of the underlying system. We focused on a traditional recommender system context, the rating prediction on the MovieLens20M movie dataset and the Goodreads book dataset. We input link stream features along with some content-based ones into a gradient boosting machine (XGBoost) and show that it outperforms significantly a sole content-based solution. These encouraging results call for further exploration of this original modelling and its integration to complete state-of-the-art recommender systems algorithms. Link streams and graphs, as natural visualizations of recom-mender systems, may offer more interpretability in a time when algorithm transparency is an increasingly important topic of discussion. We also hope that these results will sparkle interesting discussions in the community about the connections between link streams and traditional methods (matrix fac-torization, deep learning).

International audience

International audience; A novel framework called Graph Diffusion & PCA (GDPCA) is proposed in the context of semi-supervised learning on graph structured data. It combines a modified Principal Component Analysis with the classical supervised loss and Laplacian regularization, thus handling the case where the adjacency matrix is sparse and avoiding the curse of dimensionality. Our framework can be applied to non-graph datasets as well, such as images by constructing similarity graph. GDPCA improves node classification by enriching the local graph structure by node covariance. We demonstrate the performance of GDPCA in experiments on citation networks and images, and we show that GDPCA compares favourably with the best state-of-the-art algorithms and has significantly lower computational complexity.

CBMI 2018 was organized in cooperation with the IEEE Signal Processing Society and the European Association for Signal Processing (EURASIP); International audience

International audience; Continuous advancements in Information and Communication Technology and the emergence of the Big Data era have altered how traditional power systems function. Such developments have led to increased reliability and efficiency, in turn contributing to operational, economic, and environmental improvements and leading to the development of a new technique known as Demand Side Management or DSM. In essence, DSM is a management activity that encourages users to optimize their electricity consumption by controlling the operation of their electrical appliances to reduce utility bills and their use during peak times. While users may save money on electricity costs by rescheduling their power consumption, they may also experience inconvenience due to the inflexibility of getting power on demand. Hence, several challenges must be considered to achieve a successful DSM. In this work, we analyze the power scheduling techniques in Smart Houses as proposed in most cited papers. We then examine the advantages and drawbacks of such methods and compare their contributions based on operational, economic, and environmental aspects.

As Machine Learning (ML) is now widely applied in many domains, in both research and industry, an understanding of what is happening inside the black box is becoming a growing demand, especially by non-experts of these models. Several approaches had thus been developed to provide clear insights of a model prediction for a particular observation but at the cost of long computation time or restrictive hypothesis that does not fully take into account interaction between attributes. This paper provides methods based on the detection of relevant groups of attributes -- named coalitions -- influencing a prediction and compares them with the literature. Our results show that these coalitional methods are more efficient than existing ones such as SHapley Additive exPlanation (SHAP). Computation time is shortened while preserving an acceptable accuracy of individual prediction explanations. Therefore, this enables wider practical use of explanation methods to increase trust between developed ML models, end-users, and whoever impacted by any decision where these models played a role., Paper submitted to Information Systems Frontiers (Special Issue of the ADBIS 2020 conference)

International audience; Providing recommendation to agents (e.g. people or organizations) regarding whom they should collaborate with in order to reach some objective is a recurring problem in a wide range of domains. It can be useful for instance in the context of collaborative machine learning, grouped purchases, and group holidays. This problem has been modeled by hedonic games, but this generic formulation cannot easily be used to provide efficient algorithmic solutions. In this work, we define a class of hedonist games that allows us to provide an algorithmic solution to the collaboration recommendation problem by means of a clustering algorithm. We evaluate our algorithm, theoretically and experimentally and show that it performs better than other clustering algorithms in this context.

International audience; The influence of authors is mostly based on their capacity to form specific self-contained and/or active research communities or topics while also inspiring fruitful spin-off research derived from those communities or topics. Accurately estimating author influence and its indirect effect in inspiring external research creativity must thus be based on very precise community roles as well as topic-role analysis on citing material. In this paper we thus implement, compare, and combine two different kinds of knowledge mapping approaches used to track authors’ influence over time—a recent topic-based mapping approach and an original hybrid community-based mapping approach. The experimental context of our study is the analysis of the dynamics and of the influence of the research of Prof. Liu Zeyuan, the most important contributor in the field of Science of Science in China. This analysis focuses on papers citing Professor Liu Zeyuan's research and highlights the full extent of his creative thinking.

International audience; Named entity processing over historical texts is more and more being used due to the massive documents and archives being stored in digital libraries. However, due to the poor annotated resources of historical nature, information extraction performances fall behind those on contemporary texts. In this paper, we introduce the development of the NewsEye resource, a multilingual dataset for named entity recognition and linking enriched with stances towards named entities. The dataset is comprised of diachronic historical newspaper material published between 1850 and 1950 in French, German, Finnish, and Swedish. Such historical resource is essential in the context of developing and evaluating named entity processing systems. It evenly allows enhancing the performances of existing approaches on historical documents which enables adequate and efficient semantic indexing of historical documents on digital cultural heritage collections.

In the Web of Things (WoT) context, an increasing number of objects provide functions as RESTful services (resources), that can be composed with other existing resources, to create value-added processes (compositions). However, to form a composition, selecting the suitable resources is becoming more challenging, due to: (1) the growing number of resources providing identical functions, which calls for the use of Quality of Resource (QoR) to distinguish between them, and (2) the transient nature of resource availability as a result of objects’ sporadic connectivity in the WoT environments. In this chapter, we present a QoR-driven resource selection approach that forms i-compositions (with i \(\in \mathbb {N}^{*}\)) offering different implementation alternatives. This is done using a selection strategy adaptor that considers QoR constraints and Inputs/Outputs matching of related resources, as well as resource availability and users’ different needs (e.g., optimal compositions having the highest scores, and optimistic compositions having acceptable scores but obtained in more satisfactory delays). Analysis are made to evaluate our resource quality model against existing ones, and experiments are conducted in different environments setups to study the performance of our work.

International audience; Deep learning models have strong abilities in learning features and they have been successfully applied in hyperspectral images (HSIs). However, the training of most deep learning models requires labeled samples and the collection of labeled samples are labor-consuming in HSI. In addition, single-level features from a single layer are usually considered, which may result in the loss of some important information. Using multiple networks to obtain multi-level features is a solution, but at the cost of longer training time and computational complexity. To solve these problems, a novel unsupervised multi-level feature extraction framework that is based on a three dimensional convolutional autoencoder (3D-CAE) is proposed in this paper. The designed 3D-CAE is stacked by fully 3D convolutional layers and 3D deconvolutional layers, which allows for the spectral-spatial information of targets to be mined simultaneously. Besides, the 3D-CAE can be trained in an unsupervised way without involving labeled samples. Moreover, the multi-level features are directly obtained from the encoded layers with different scales and resolutions, which is more efficient than using multiple networks to get them. The effectiveness of the proposed multi-level features is verified on two hyperspectral data sets. The results demonstrate that the proposed method has great promise in unsupervised feature learning and can help us to further improve the hyperspectral classification when compared with single-level features.

International audience; A key issue for Twitter users relates to the summarization of the continuous and overwhelming stream of information. Many approaches for tweet summarization were proposed in the literature. It is however difficult to compare them given the lack of standard and accessible test collection. This absence can be due to the efforts needed to construct such a (large) dataset. In this paper, we propose to capitalize on the dataset proposed for the TREC Incident Streams track, which was not intended to evaluate automatic summarization. We show why and how this dataset is usable for this purpose, focusing on extractive summarization. Indeed, when producing additional annotations on a subset of the TREC Incident Streams (IS) dataset with particular initial assessors' annotations, it appears to respect the criteria identified in the literature for automatic summarization. For this, we studied the original TREC IS dataset and then proposed a subset summarizing each event, based on the initial assessors' annotations. This subset is evaluated according to the criteria previously mentioned. Several widely used state-of-the-art models for automatic text summarization, some specific to tweets and some adapted to tweet summarization, were finally tested on the proposed dataset. For easy reproducibility, the code used to build the dataset, our additional annotations, and the experiments made on the dataset are provided on our Github.

International audience; This article questions the effects on urban research dynamics of the Big Data and AI turn in urban management. Increasing access to large datasets collected in real time could make certain mathematical models developed in research fields related to the management of urban systems obsolete. These ongoing evolutions are the subject of numerous works whose main angle of reflection is the future of cities rather than the transformations at work in the academic field. Our article proposes grasp the scientific dynamics in areas of research related to two urban systems: transportation and water. The article demonstrates the importance of grasping these dynamics if we want to be able to apprehend what the urban management of tomorrow's cities will be like. To analyse these research areas’ dynamics, we use two complementary materials: bibliometric data and interviews. The interviews conducted in 2018 with academics and higher education officials in Paris and Edinburgh suggest avenues for hybridization between traditional modelling approaches and research in machine learning, artificial intelligence and big data. The bibliometric analysis highlight the trends at work: it shows that traffic flow as well as transportation studies are focussing more and more on AI and Big Data and that traffic flow studies are arousing a growing interest among computer scientists while, so far, this interest is less pronounced in the water research area, and more especially regarding water quality. The differences observed between research on transportation and that on water confirm the multifaceted nature of the developments at work and encourage us to reject overly hasty and simplistic generalisations about the transformations underway.; Cet article interroge les effets sur la recherche urbaine de l'avènement des Big Data et de l'IA dans la gestion urbaine. L'accès croissant à de grands ensembles de données collectées en temps réel peut-il rendre obsolètes certains modèles mathématiques développés dans les domaines de recherche liés à la gestion des systèmes urbains ? Cette question fait l'objet de nombreux travaux dont l'angle principal est le futur des villes plutôt que celui des transformations à l'œuvre dans le monde académique. Notre article propose d'aborder cette question sous l'angle des dynamiques internes des champs scientifiques liés à deux systèmes urbains : le transport et l'eau. Il démontre l'importance de saisir ces dynamiques si l'on veut pouvoir appréhender ce que sera la gestion urbaine des villes de demain. Pour les analyser, nous utilisons deux matériaux complémentaires : des données bibliométriques et des entretiens. Les entretiens réalisés, en 2018, avec des universitaires et des responsables de l'enseignement supérieur à Paris et à Édimbourg suggèrent des pistes d'hybridation entre les approches traditionnelles de modélisation et la recherche en apprentissage, intelligence artificielle et Big Data. L'analyse bibliométrique met en évidence les tendances suivantes : elle montre que les études sur les transports et sur le trafic en particulier mobilisent de plus en plus l'IA et les Big Data et que le trafic suscite un intérêt croissant chez les informaticiens tandis que, jusqu'à présent, cet intérêt est moins prononcé dans le domaine de la recherche sur l'eau, et plus particulièrement sur la qualité de l'eau. Les différences observées entre la recherche sur les transports et celle sur l'eau confirment la nature multiforme des développements à l'œuvre et nous incitent à rejeter les généralisations trop hâtives et simplistes sur les transformations en cours.

International audience; Despite a huge leap in performance of face recognition systems in recent years, some cases remain challenging for them while being trivial for humans. This is because a human brain is exploiting much more information than the face appearance to identify a person. In this work, we aim at capturing the social context of unlabeled observed faces in order to improve face retrieval. In particular, we propose a framework that substantially improves face retrieval by exploiting the faces occurring simultaneously in a query's context to infer a multi-dimensional social context descriptor. Combining this compact structural descriptor with the individual visual face features in a common feature vector considerably increases the correct face retrieval rate and allows to disambiguate a large proportion of query results of different persons that are barely distinguishable visually. To evaluate our framework, we also introduce a new large dataset of faces of French TV personalities organised in TV shows in order to capture the co-occurrence relations between people. On this dataset, our framework is able to improve the mean Average Precision over a set of internal queries from 67.93% (using only facial features extracted with a state-of-the-art pre-trained model) to 78.16% (using both facial features and faces co-occurrences), and from 67.88% to 77.36% over a set of external queries.

International audience; Optical character recognition (OCR) is one of the most popular techniques used for converting printed documents into machine-readable ones. While OCR engines can do well with modern text, their performance is unfortunately significantly reduced on historical materials. Additionally, many texts have already been processed by various out-of-date digitisation techniques. As a consequence, digitised texts are noisy and need to be post-corrected. This article clarifies the importance of enhancing quality of OCR results by studying their effects on information retrieval and natural language processing applications. We then define the post-OCR processing problem, illustrate its typical pipeline, and review the state-of-the-art post-OCR processing approaches. Evaluation metrics, accessible datasets, language resources, and useful toolkits are also reported. Furthermore, the work identifies the current trend and outlines some research directions of this field.

International audience; Event detection is one of the most important research topics in social media analysis. Despite this interest, few researchers have addressed the problem of identifying geolocated events in an unsupervised way, and none includes user interests during the process. In this paper, we tackle the problem of local event detection from social media data. We present a method to automatically identify events by evaluating the burstiness of hashtags in a geographical area and a time interval, and at the same time integrating user feedback. We devise two algorithms to discover user-driven events. The first one relies on an exact enumeration process, while the other directly samples the space of events. In our empirical study, we provide evidence that geolocated events cannot be detected by non location-aware methods. We also show that our methods (i) outperform by a factor of two to several orders of magnitude state-of-the-art methods designed to discover geolocated events, (ii) are more robust to noise, (iii) and produce high quality events with respect to user interests.

International audience; Several deep neural ranking models have been proposed in the recent IR literature. While their transferability to one target domain held by a dataset has been widely addressed using traditional domain adaptation strategies, the question of their cross-domain transferability is still under-studied. We study here in what extent neural ranking models catastrophically forget old knowledge acquired from previously observed domains after acquiring new knowledge, leading to performance decrease on those domains. Our experiments show that the effectiveness of neural IR ranking models is achieved at the cost of catastrophic forgetting and that a lifelong learning strategy using a cross-domain regularizer successfully mitigates the problem. Using an explanatory approach built on a regression model, we also show the effect of domain characteristics on the rise of catastrophic forgetting. We believe that the obtained results can be useful for both theoretical and practical future work in neural IR.