Section: New Results

The Mining of Complex Data

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

Keywords:

formal concept analysis, relational concept analysis, pattern structures, pattern mining, association rule, graph mining, sequence mining, biclustering

Formal Concept Analysis and pattern mining are suitable symbolic methods for KDDK, that may be used for real-sized applications. Global improvements are 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 extending these symbolic data mining methods for working on complex data (e.g. textual documents, biological, chemical or medical data), involving objects with multi-valued attributes (e.g. domains or intervals), n-ary relations, sequences, trees and graphs.

FCA and Variations: RCA, Pattern Structures and Biclustering

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 [2] [131] . 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 can play has an important role in KDDK, especially in text mining [105] .

Another extension of FCA is based on Pattern Structures (PS) [112] , which allows to build a concept lattice from complex data, e.g. nominal, numerical, and interval data [119] . Since then, we worked on some experiments involving pattern structures, namely sequence mining [107] , information retrieval and recommendation [58] , [22] , functional dependencies [50] , [17] and biclustering [69] , [41] . One of the next step is the adaptation of pattern structures to graph mining.

Moreover, the notion of similarity between objects is also closely related to pattern structures [102] : 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). Combination of similarity and pattern structures is also under study, in particular for solving information retrieval and annotation problems.

In pattern mining as in FCA, one main problem is the volume of the output. One general idea is to extract patterns which show a “good behavior” w.r.t. a given measure. Such patterns or concepts are expected to have good characteristics and to provide effective knowledge. We have conducted in the framework of FCA a series of experiments on the so-called “stability measure”, showing that this measure is able to detect significant patterns [54] , [53] .

Finally, there is also an on-going work relating FCA and semantic web. This work focuses on the classification within a concept lattice of the answers returned by SPARQL queries. The concept lattice is then used as an index for navigating and ranking the answers w.r.t. their content and interest for a given objective [47] .

Sequence Mining

Sequence data is widely used in many applications. Consequently, mining sequential patterns and other types of knowledge from sequence data became an important data mining task. In the team, the main emphasis is on developing efficient mining algorithms for pattern classification problems. The most frequent sequences generally provide trivial information. When analyzing the set of frequent sequences with a low minimum support, the user is overwhelmed by millions of patterns.

In our recent work, we studied the notion of $\delta$-freeness for sequences. While this notion has extensively been discussed for itemsets, our work is the first to extend it to sequences. We defined an efficient algorithm devoted to the extraction of $\delta$-free sequential patterns. We presented the advantage of the $\delta$-free sequences and highlighted their importance when building sequence classifiers, and we showed how they can be used to address the feature selection problem in statistical classifiers which optimizes both accuracy and earliness of predictions [68] .

Mining and Understanding Healthcare Trajectories

With the increasing burden of chronic illnesses, administrative health care databases hold valuable information that could be used to monitor and assess the processes shaping the trajectory of care of chronic patients. In this context, temporal data mining methods are promising tools, though lacking flexibility in addressing the complex nature of medical events. In the thesis work of Elias Egho [15] , new algorithms were designed to extract patient trajectory patterns with different levels of granularity by relying on external taxonomies [62] , [34] . The algorithms rely on the general FCA framework to formalize the general notion of multidimensional healthcare trajectories. There was also another work focusing on the similarity measure among sequences. An efficient and original similarity measure was design for that purpose [8] .

Video Game Analytics

The video game industry has grown enormously over the last twenty years, bringing new challenges to the artificial intelligence and data analysis communities. We tackled this year the problem of automatic discovery of strategies in real-time strategy games through pattern mining. Such patterns are the basic units for many tasks such as automated agent design, but also to build tools for the professionally played video games in the electronic sports scene. We presented a new formalism within a sequential pattern mining approach and a novel measure, the balance measure, telling how a strategy is likely to win [51] . We experimented our methodology on a real-time strategy game that is professionally played in the electronic sport community and laid plans on a future collaboration with the MIT Game Lab.

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 based on FCA and RCA for ontology engineering from heterogeneous ontological resources. This methodology is based on both FCA and RCA, and was previously successfully applied in domains such as astronomy and biology.

In the framework of the ANR Hybride project (see  8.2.1.2 ), an engineer is implementing a robust system based on these previous research results, for preparing the way to new research directions involving trees and graphs. Moreover, we led a first successful experiment on extracting drug-drug interactions applying “lazy pattern structure classification” to syntactic trees. In addition, in his thesis work, Mohsen Sayed focused on extracting relations between named entities using graph mining methods applied to dependency graphs [67] . We are currently investigating how this approach can be generalized, i.e. how to detect a relation between complex expressions which are not previously recognized as named entities.

The notion of “Jumping Emerging Patterns” (JEP) previously used in chemistry [101] , was updated and adapted in the context of text mining within the ANR Termith project. The objective is to design a learning method for filtering candidate terms within a full text and to decide whether an occurrence should be tagged as a term, i.e. a positive example, or as a simple word, i.e. a negative example. The method extracts from a training set all JEPs which are considered as hypotheses. To reduce the number of JEPs and to retain only the more significant JEPs from a linguistic point of view, JEPs are weighted and a constraint solver is used to verify the maximal coverage of the positive examples. Results are currently under evaluation.