Cheikh Talibouya Diop

Composition of Mining Contexts for Efficient Extraction of Association Rules

Association rule mining often requires the repeated execution of some extraction algorithm for different values of the support and confidence thresholds, as well as for different source datasets. This is an expensive process, even if we use the best existing algorithms. Hence the need for incremental mining, whereby mining results already obtained can be used to accelerate subsequent steps in the mining process.In this paper, we present an approach for the incremental mining of multi-dimensional association rules. In our approach, association rule mining takes place in a mining context which specifies the form of rules to be mined. Incremental mining is obtained by combining mining contexts using relational algebra operations.

Condensed Representations for Sets of Mining Queries

In this paper, we propose a general framework for condensed representations of sets of mining queries. To this end, we adapt the standard notions of maximal, closed and key patterns introduced in previous works, including those dealing with condensed representations. Whereas these previous works concentrate on condensed representations of the answer to a single mining query, we consider the more general case of sets of mining queries defined by monotonic and anti-monotonic selection predicates.

Mining contextual preference rules for building user profiles

The emerging of ubiquitous computing technologies in recent years has given rise to a new field of research consisting in incorporating context-aware preference querying facilities in database systems. One important step in this setting is the Preference Elicitation task which consists in providing the user ways to inform his/her choice on pairs of objects with a minimal effort. In this paper we propose an automatic preference elicitation method based on mining techniques. The method consists in extracting a user profile from a set of user preference samples. In our setting, a profile is specified by a set of contextual preference rules verifying properties of soundness and conciseness. We evaluate the efficacy of the proposed method in a series of experiments executed on a real-world database of user preferences about movies.

Contextual preference mining for user profile construction

The emerging of ubiquitous computing technologies in recent years has given rise to a new field of research consisting in incorporating context-aware preference querying facilities in database systems. One important step in this setting is the Preference Elicitation task which consists in providing the user ways to inform his/her choice on pairs of objects with a minimal effort. In this paper we propose an automatic preference elicitation method based on mining techniques. The method consists in extracting a user profile from a set of user preference samples. In our setting, a profile is specified by a set of contextual preference rules verifying properties of soundness and conciseness. After proving that the problem is NP-complete, we propose a resolution in 2 phases. The first phase extracts all individual user preferences by means of contextual preference rules. The second phase builds the user profile starting from this collection of rules using a greedy method. To assess the quality of user profiles, we propose three ranking techniques benefiting from these profiles that enable us to rank objects according to user preferences. We evaluate the efficacy of our three ranking strategies and compare them with a well-known ranking method (SVMRank). The evaluation is carried out through an extensive set of experiments executed on a real-world database of user preferences about movies. HighlightsWe propose a new method for mining user profiles from preference databases.Our method is the first one based on pattern mining techniques.It builds readable user profile based on the notion of contextual preference rules.Three ranking techniques are proposed to rank objects according to a user profile.A set of experiments on a real-world database showed the efficiency of the method.

Cube based summaries of large association rule sets

A major problem when dealing with association rules postprocessing is the huge amount of extracted rules. Several approaches have been implemented to summarize them. However, the obtained summaries are generally difficult to analyse because they suffer from the lack of navigational tools. In this paper, we propose a novel method for summarizing large sets of association rules. Our approach enables to obtain from a rule set, several summaries called Cube Based Summaries (CBSs). We show that the CBSs can be represented as cubes and we give an overview of OLAP 1 navigational operations that can be used to explore them. Moreover, we define a new quality measure called homogeneity, to evaluate the interestingness of CBSs. Finally, we propose an algorithm that generates a relevant CBS w.r.t. a quality measure, to initialize the exploration. The evaluation of our algorithm on benchmarks proves the effectiveness of our approach.

Computation of mining queries: an algebraic approach

Mining frequent queries often requires the repeated execution of some extraction algorithm for different values of the support, as well as for different source datasets. This is an expensive process, even if we use the best existing algorithms. Hence the need for iterative mining, whereby mining results already obtained are re-used to accelerate subsequent steps in the mining process. In this paper, we present an approach for the iterative mining of frequent queries. Our approach is based on the notion of mining context, where a mining context is a set of queries over the same schema. We define operations on mining contexts, based on the standard relational algebra, and we also introduce new operators, one of which for computing frequent queries. We first study the properties of the operators, then we consider particular mining contexts using biases for which frequent queries can be computed using any level-wise algorithm. Iterative mining is obtained by combining these particular contexts using our set of operations. We have implemented our approach and conducted experiments that show its efficiency in mining frequent queries.

Extraction itérative de requêtes fréquentes

Most approaches to knowledge discovery in databases consider extractions one at a time. However it is well known that, in practice, any extraction process is interactive and iterative. In this paper, we introduce an approach that can take benefit from previously computed extractions in the computation of the current extraction. To do so, we assume that the answers to previous extractions are stored and we use properties of query containment for pruning candidate queries. First, we introduce our approach in the context of deductive databases. Then, we show that restricting our framework to the case of relational queries allows for reducing the complexity of query containment tests. The experimental results reported in this paper show a significant reduction of computation time when taking into account the iterative aspects mentioned above.