Catálogo de publicaciones - libros

Among the family of the local patterns, episodes are commonly used when mining a single or multiple sequences of discrete events. An episode reflects a qualitative relation over event types, and the refinement of episodes to incorporate quantitative temporal information is still an on going research, with many application opportunities. In this paper, focusing on serial episodes, we design such a refinement called and give a corresponding extraction algorithm. The three most salient features of these quantitative episodes are: (1) their ability to characterize main groups of homogeneous behaviors among the occurrences, according to the duration of the steps, and providing quantitative bounds of these durations organized in a tree structure; (2) the possibility to extract them in a complete way; and (3) to perform such extractions at the cost of a limited overhead with respect to the extraction of standard episodes.

The overall goal of this paper is to devise a flexible and declarative query language for specifying or describing particular knowledge discovery scenarios. We introduce one such language, called IQL. IQL is intended as a general, descriptive, declarative, extendable and implementable language for inductive querying that supports the mining of both local and global patterns, reasoning about inductive queries and query processing using logic, as well as the flexible incorporation of new primitives and solvers. IQL is an extension of the tuple relational calculus that includes functions as primitives. The language integrates ideas from several other declarative programming languages, such as pattern matching and function typing. We hope that it will be useful as an overall specification language for integrating data mining systems and principles.

Missing values issue in databases is an important problem because missing values bias the information provided by the usual data mining methods. In this paper, we are searching for mining patterns satisfying correct properties in presence of missing values (it means that these patterns must satisfy the properties in the corresponding complete database). We focus on -free patterns. Thanks to a new definition of this property suitable for incomplete data and compatible with the usual one, we certify that the extracted -free patterns in an incomplete database also satisfy this property in the corresponding complete database. Moreover, this approach enables to provide an anti-monotone criterion with respect to the pattern inclusion and thus design an efficient level-wise algorithm which extracts correct -free patterns in presence of missing values.

Mining patterns under many kinds of constraints is a key point to successfully get new knowledge. In this paper, we propose an efficient new algorithm which soundly and completely mines patterns with various constraints from large data and takes into account external data represented by several heterogeneous datasets. Constraints are freely built of a large set of primitives and enable to link the information scattered in various knowledge sources. Efficiency is achieved thanks to a new closure operator providing an interval pruning strategy applied during the depth-first search of a pattern space. A transcriptomic case study shows the effectiveness and scalability of our approach. It also demonstrates a way to employ background knowledge, such as free texts or gene ontologies, in the discovery of meaningful patterns.

Frequent itemset mining is often regarded as advanced querying where a user specifies the source dataset and pattern constraints using a given constraint model. Recently, a new problem of optimizing processing of sets of frequent itemset queries has been considered and two multiple query optimization techniques for frequent itemset queries: Mine Merge and Common Counting have been proposed and tested on the Apriori algorithm. In this paper we discuss and experimentally evaluate three strategies for concurrent processing of frequent itemset queries using FP-growth as a basic frequent itemset mining algorithm. The first strategy is Mine Merge, which does not depend on a particular mining algorithm and can be applied to FP-growth without modifications. The second is an implementation of the general idea of Common Counting for FP-growth. The last is a completely new strategy, motivated by identified shortcomings of the previous two strategies in the context of FP-growth.

In this paper, we address the ambitious task of formulating a general framework for data mining. We discuss the requirements that such a framework should fulfill: It should elegantly handle different types of data, different data mining tasks, and different types of patterns/models. We also discuss data mining languages and what they should support: this includes the design and implementation of data mining algorithms, as well as their composition into nontrivial multi-step knowledge discovery scenarios relevant for practical application. We proceed by laying out some basic concepts, starting with (structured) data and generalizations (e.g., patterns and models) and continuing with data mining tasks and basic components of data mining algorithms (i.e., refinement operators, distances, features and kernels). We next discuss how to use these concepts to formulate constraint-based data mining tasks and design generic data mining algorithms. We finally discuss how these components would fit in the overall framework and in particular into a language for data mining and knowledge discovery.