Section: New Results

The Mining of Complex Data

Participants : Mehwish Alam, Thomas Bourquard, Aleksey Buzmakov, Victor Codocedo, Adrien Coulet, Elias Egho, Nicolas Jay, Florence Le Ber, Ioanna Lykourentzou, Luis Felipe Melo, Amedeo Napoli, Chedy Raïssi, My Thao Tang, Yannick Toussaint.

formal concept analysis, relational concept analysis, pattern structures, search for frequent itemsets, association rule extraction, mining of complex data, graph mining, skylines, sequence mining, FCA in spatial and temporal reasoning

Formal concept analysis, together with itemset search and association rule extraction, are suitable symbolic methods for KDDK, that may be used for real-sized applications. Global improvements may be carried on the scope of applicability, the ease of use, the efficiency of the methods, and on the ability to fit evolving situations. Accordingly, the team is working on extensions of such symbolic data mining methods to be applied on complex data such as biological or chemical data or textual documents, involving objects with multi-valued attributes (e.g. domains or intervals), n-ary relations, sequences, trees and graphs.

FCA, RCA, and Pattern Structures

Recent advances in data and knowledge engineering have emphasized the need for Formal Concept Analysis (FCA) tools taking into account structured data. There are a few extensions of FCA for handling contexts involving complex data formats, e.g. graphs or relational data. Among them, Relational Concept Analysis (RCA) is a process for analyzing objects described both by binary and relational attributes [116] . The RCA process takes as input a collection of contexts and of inter-context relations, and yields a set of lattices, one per context, whose concepts are linked by relations. RCA has an important role in KDDK, especially in text mining [86] , [85] .

Another extension of FCA is based on Pattern Structures (PS) [94] , which allows to build a concept lattice from complex data, e.g. nominal, numerical, and interval data. In [101] ), pattern structures are used for building a concept lattice from intervals, in full compliance with FCA, thus benefiting of the efficiency of FCA algorithms. Actually, the notion of similarity between objects is closely related to these extensions of FCA: two objects are similar as soon as they share the same attributes (binary case) or attributes with similar values or the same description (at least in part). Various results were obtained in the study of the relations existing between FCA with an embedded explicit similarity measure and FCA with pattern structures [100] . Moreover, similarity is not a transitive relation and this lead us to the study of tolerance relations. In addition, a new research perspective is aimed at using frequent itemset search methods for mining interval-based data being guided by pattern structures and biclustering as well.

Advances in FCA and Pattern Mining

In the context of environmental sciences, research work is in concern with the mining of complex hydroecological data with concept lattices. In particular, Florence Le Ber –as a member of UMR 7517 Lhyges, Strasbourg– is the scientific head of an ANR project named “FRESQUEAU” (2011–2014) dealing with FCA and data mining and hydroecological data (see http://engees.unistra.fr/site/recherche/projets/anr-fresqueau/ ).

In this framework, concept lattices based on multi-valued contexts have been used for characterizing macroinvertebrate communities in wetland and their seasonal evolution [19] . Within the ANR Fresqueau project we are studying tools for sequential pattern extraction taking into account spatial relations [56] , [43] .

From another point of view, miscanthus is a perennial crop used for biomass production. Its implantation is rather new, and there is few farms cultivating miscanthus in France. Understanding the farmers' choices for allocating miscanthus in their farmland is a main challenge. The CBR model is investigated for modeling these choices from farm surveys, including spatial reasoning aspects [20] , [47]  [41] .

For completing the work on FCA and itemset search, there is still on-going work on frequent and rare itemset search, for being able to build lattices from very large data and completing the algorithm collection of the Coron platform. Work is still in progress on the design of an integrated and modular algorithm for searching for closed and generators itemsets, and equivalence classes of itemsets, thus enabling the construction of the associated lattice [121] . This research aspect is also linked to the research carried on within a the PICS CaDoE research project (see Section  8.1.1.3 ). In addition, there is also research work carried on different aspects involving the management of big data in the context of the BioIntelligence Project and the Quaero Project.

Skylines, sequential data, privacy and E-sports analytics

Pattern discovery is at the core of numerous data mining tasks. Although many methods focus on efficiency in pattern mining, they still suffer from the problem of choosing a threshold that influences the final extraction result. One goal is to make the results of pattern mining useful from a user-preference point of view. That is, take into account some domain knowledge to guide the pattern mining process. To this end, we integrate into the pattern discovery process the idea of skyline queries in order to mine skyline patterns in a threshold-free manner. This forms the basis for a novel approach to mining skyline patterns. The efficiency of our approach was illustrated over a use case from chemoinformatics and we showed that small sets of dominant patterns are produced under various measures that are interesting for chemical engineers and researchers.

Sequence data is widely used in many applications. Consequently, mining sequential patterns and other types of knowledge from sequence data has become an important data mining task. The main emphasis has been on developing efficient mining algorithms and effective pattern representation.

However, important fundamental problems still remained open: $\left(i\right)$ given a sequence database, can we have an upper bound on the number of sequential patterns in the database? $\left(ii\right)$ Is the efficiency of the sequence classifier only based on accuracy? $\left(iii\right)$ Do the classifiers need the entire set of extracted patterns or a smaller set with the same expressiveness power?

In the field of the management of sequential date in medicine, analysis of health care trajectories led to the development of a new sequential pattern mining method [42] . The MMISP algorithm is able to efficiently extract sequential patterns composed of itemsets and multidimensional items. The multidimensional items can be described with additional taxonomic knowledge, allowing mining with appropriate levels of granularity. In parallel, a new measure has been created to compute the similarity between sequences of itemsets [78] .

Orpailleur is one of the few project-teams working on privacy challenges which are becoming a core issue with different scientific problems in computer science. With technology infiltrating more and more every aspect of our lives, each human activity leaves a digital trace in some repository. Vast amounts of personal data are implicitly or explicitly created each day, and rarely one is aware of the extent of information that is kept, processed and analyzed without his knowledge or consent. These personal data give rise to significant concerns about user privacy, since important and sensitive details about private life are collected and exploited by third parties. The goal of privacy preservation technologies is to provide tools that allow greater control over the dissemination of user data. A promising trend in the field is Privacy Preserving Data Publishing (PPDP), which allows sharing of anonymized data. Anonymizing a dataset is not limited to the removal of direct identifiers that might exist in a dataset, e.g. the full name or the Social Security Number of a person. It also includes removing secondary information, e.g. like age, zip code that might lead indirectly to the true identity of an individual.

Existing research on this problem either perturbs the data, publishes them in disjoint groups disassociated from their sensitive labels, or generalizes their values by assuming the availability of a generalization hierarchy. In a recent work, we proposed a novel alternative [54] . Our publication method also puts data in a generalized form, but does not require that published records form disjoint groups and does not assume a hierarchy either. Instead, it employs generalized bitmaps and recasts data values in a nonreciprocal manner.

One of the most fascinating challenges of our time is understanding the complexity of the global interconnected society we inhabit. Today we have the opportunity to observe and measure how our society intimately works, by analyzing the big data. i.e, the digital breadcrumbs of human activities sensed as a by-product of the ICT systems that we use. These data describe the daily human activities: for instance, automated payment systems record the tracks of our purchases, search engines record the logs of our queries for finding information on the web, social networking services record our connections to friends, colleagues and collaborators, wireless networks and mobile devices record the traces of our movements and our communications. These social data are at the heart of the idea of a knowledge society, where decisions can be taken on the basis of knowledge in these data.

Social network data analysis raises concerns about the privacy of related entities or individuals. We theoretically establish that any kind of structural identification attack can effectively be prevented using random edge perturbation and show that, surprisingly, important properties of the whole network, as well as of subgraphs thereof, can be accurately calculated and hence data analysis tasks performed on the perturbed data, given that the legitimate data recipient knows the perturbation probability as well [53] .

"Electronic-sport" (E-Sport) is now established as a new entertainment genre. More and more players enjoy streaming their games, which attract even more viewers. In fact, in a recent social study, casual players were found to prefer watching professional gamers rather than playing the game themselves. Within this context, advertising provides a significant source of revenue to the professional players, the casters (displaying other people's games) and the game streaming platforms. In a recent work with Mehdi Kaytoue, we started focusing on the huge amount of data generated by electronic games. We crawled, during more than 100 days, the most popular among such specialized platforms: Twitch.tv.

Thanks to these gigabytes of data, we proposed a first characterization of a new Web community, and we showed, among other results, that the number of viewers of a streaming session evolves in a predictable way, that audience peaks of a game are explainable and that a Condorcet method can be used to sensibly rank the streamers by popularity [45] . This work should bring to light the study of E-Sport and its growing community for computer scientists and sociologists. They indeed deserve the attention of industrial partners (for the large amount of money involved) and researchers (for interesting problems in social network dynamics, personalized recommendation, sentiment analysis, etc.).

KDDK in Text Mining

Ontologies help software and human agents to communicate by providing shared and common domain knowledge, and by supporting various tasks, e.g. problem-solving and information retrieval. In practice, building an ontology depends on a number of “ontological resources” having different types: thesaurus, dictionaries, texts, databases, and ontologies themselves. We are currently working on the design of a methodology and the implementation of a system for ontology engineering from heterogeneous ontological resources. This methodology is based on both FCA and RCA, and was previously successfully applied in contexts such as astronomy and biology. At present, an engineer is implementing a robust system being guided by the previous research results and preparing the way for some new research directions involving trees and graphs (see also the work on the ANR Hybride project).