L'évolution rapide des techniques de génération et de stockage de données a permis à de nombreux organismes la création de bases de données volumineuses, pour stocker l'information nécessaire à leurs activités. Ces bases de données qui deviennent de plus en plus importantes sont réellement peu exploitées, alors qu'elles cachent des connaissances potentiellement utiles pour l'organisation. L'extraction de ces informations enfouies dans ces masses de données est traitée par la fouille de données ("Data Mining"). Ce projet de mémoire traite plus particulièrement le problème d'extraction des informations sous forme de règles d'associations. Le problème de la pertinence et de l'utilité des règles extraites est un problème majeur de l'extraction des règles d'associations. Ce problème est lié au nombre important de règles extraites et à la présence d'une forte proportion de règles redondantes. Nombreuses techniques de réduction de la famille de règles ont été publiées. Dans ce contexte, les résultats obtenus par l'analyse formelle des concepts (AFC) ont permis de définir un sous-ensemble de l'ensemble des règles d'associations valides appelés bases informatives. La génération de ces bases informatives se fait par une extraction efficace des itemsets fermés fréquents et leurs générateurs minimaux associés. Les générateurs minimaux composent les prémisses minimales de ces règles alors que leurs fermetures composent les conclusions maximales de ces règles. Cependant un survol de la littérature montre que les générateurs minimaux composant l'antécédent et la conséquence de ces bases, contiennent encore de la redondance. Une représentation réduite de ces générateurs minimaux est utile pour révéler la relation d'équivalence parmi les générateurs minimaux. Une étude a été menée dernièrement dans ce sens dans laquelle l'algorithme DSFS_MINER a été proposé et validé, permettant l'extraction d'une représentation succincte sans perte d'informations des générateurs minimaux. Notre contribution dans ce projet réside d'une part, dans l'étude et l'expérimentation d'approches de représentations succinctes des générateurs minimaux, et d'autre part, dans la proposition d'un algorithme d'expansion permettant la dérivation de tous les générateurs minimaux afin de constituer la famille entière des générateurs minimaux du contexte d'extraction. ______________________________________________________________________________ MOTS-CLÉS DE L’AUTEUR : Data Mining, Règles d'associations, Analyse formelle des concepts, Générateurs minimaux, Itemset fermés, Générateur minimal, Représentation succincte des générateurs minimaux.

A problem that has been the focus of much recent research in privacy preserving data-mining is the frequent itemset hiding (FIH) problem. Identifying itemsets that appear together frequently in customer transactions is a common task in association rule mining. Organizations that share data with business partners may consider some of the frequent itemsets sensitive and aim to hide such sensitive itemsets by removing items from certain transactions. Since such modifications adversely affect the utility of the database for data mining applications, the goal is to remove as few items as possible. Since the frequent itemset hiding problem is NP-hard and practical instances of this problem are too large to be solved optimally, there is a need for heuristic methods that provide good solutions. This dissertation developed a new method called Min_Items_Removed, using the Frequent Pattern Tree (FP-Tree) that outperforms extant methods for the FIH problem. The FP-Tree enables the compression of large databases into significantly smaller data structures. As a result of this compression, a search may be performed with increased speed and efficiency. To evaluate the effectiveness and performance of the Min_Items_Removed algorithm, eight experiments were conducted. The results showed that the Min_Items_Removed algorithm yields better quality solutions than extant methods in terms of minimizing the number of removed items. In addition, the results showed that the newly introduced metric (normalized number of leaves) is a very good indicator of the problem size or difficulty of the problem instance that is independent of the number of sensitive itemsets.

Utility mining algorithms have recently been proposed to discover high utility itemsets from a quantitative database. Factors such as profits or prices are concerned in measuring the utility values of purchased items for revealing more useful knowledge to managers. Nearly all the existing algorithms are performed in a batch way to extract high utility itemsets. In real-world applications, transactions may, however, be inserted, deleted or modified in a database. The batch mining procedure requires more computational time for rescanning the whole updated database to maintain the up-to-date knowledge. In the first part of this thesis, two algorithms for data insertion and data deletion are respectively proposed for efficiently updating the discovered high utility itemsets based on pre-large concepts. The proposed algorithms firstly partition itemsets into three parts with nine cases according to whether they are large (high), pre-large or small transaction-weighted utilization in the original database. Each part is then performed by its own procedure to maintain and update the discovered high utility itemsets. Based on the pre-large concepts, the original database only need to be rescanned for much fewer itemsets in the maintenance process of high utility itemsets. Besides, the risk of privacy threats usually exists in the process of data collection and data dissemination. Sensitive or personal information are required to be kept as private information before they are shared or published. Privacy-preserving utility mining (PPUM) has thus become an important issue in recent years. In the second part of this thesis, two evolutionary privacy-preserving utility mining algorithms to hide sensitive high utility itemsets in data sanitization for inserting dummy transactions and deleting transactions are respectively proposed. The two evolutionary privacy-preserving utility mining algorithms find appropriate transactions for insertion and deletion in the data-sanitization process. They adopt a flexible evaluation function with three factors. Different weights are assigned to the three factors depending on users¡¦ preference. The maintenance algorithms proposed in the first part of this thesis are also used in the GA-based approach to reduce the cost of rescanning databases, thus speeding up the evaluation process of chromosomes. Experiments are conducted as well to evaluate the performance of the proposed algorithms.

Data mining is proving itself to be a very important field as the data available is increasing exponentially, thanks to first computerization and now internetization. On the other hand, cluster computing systems made up of commodity hardware are becoming widespread, along with the multicore processor architectures. This high computing power is synthesized with data mining to process huge amounts of data and to reach information and knowledge. Frequent itemset mining is a special subtopic of data mining because it is an integral part of many types of data mining tasks. Often this task is a prerequisite for many other data mining algorithms, most notably algorithms in the association rule mining area. For this reason, it is studied heavily in the literature. In this thesis, a parallel implementation of CLOSET+, a frequent closed itemset mining algorithm, is presented. The CLOSET+ algorithm has been modified to run on multiple processors simultaneously, in order to obtain results faster. Open MPI and Boost libraries have been used for the communication between different processes and the program has been tested on different inputs and parameters. Experimental results show that the algorithm exhibits high speedup and eficiency for dense data when the support value is higher than a determined value. Proposed parallel algorithm could prove to be useful for application areas where fast response is needed for low to medium number of frequent closed itemsets. A particular application area is the Web where online applications have similar requirements.

In binary-transaction data-mining, traditional frequent itemset mining often produces results which are not straightforward to interpret. To overcome this problem, probability models are often used to produce more compact and conclusive results, albeit with some loss of accuracy. Bayesian statistics have been widely used in the development of probability models in machine learning in recent years and these methods have many advantages, including their abilities to avoid overfitting. In this thesis, we develop two Bayesian mixture models with the Dirichlet distribution prior and the Dirichlet process (DP) prior to improve the previous non-Bayesian mixture model developed for transaction dataset mining. First, we develop a finite Bayesian mixture model by introducing conjugate priors to the model. Then, we extend this model to an infinite Bayesian mixture using a Dirichlet process prior. The Dirichlet process mixture model is a nonparametric Bayesian model which allows for the automatic determination of an appropriate number of mixture components. We implement the inference of both mixture models using two methods: a collapsed Gibbs sampling scheme and a variational approximation algorithm. Experiments in several benchmark problems have shown that both mixture models achieve better performance than a non-Bayesian mixture model. The variational algorithm is the faster of the two approaches while the Gibbs sampling method achieves a more accurate result. The Dirichlet process mixture model can automatically grow to a proper complexity for a better approximation. However, these approaches also show that mixture models underestimate the probabilities of frequent itemsets. Consequently, these models have a higher sensitivity but a lower specificity.

No description available.

No description available.

In this thesis, we focused on the construction of classification models based on association rules. Although association rules have been predominantly used for data exploration and description, the interest in using them for prediction has rapidly increased in the data mining community. In order to mine only rules that can be used for classification, we modified the well known association rule mining algorithm Apriori to handle user-defined input constraints. We considered constraints that require the presence/absence of particular items, or that limit the number of items, in the antecedents and/or the consequents of the rules. We developed a characterization of those itemsets that will potentially form rules that satisfy the given constraints. This characterization allows us to prune during itemset construction itemsets such that neither they nor any of their supersets will form valid rules. This improves the time performance of itemset construction. Using this characterization, we implemented a classification system based on association rules and compared the performance of several model construction methods, including CBA, and several model deployment modes to make predictions. Although the data mining community has dealt only with the classification of single-valued attributes, there are several domains in which the classification target is set-valued. Hence, we enhanced our classification system with a novel approach to handle the prediction of set-valued class attributes. Since the traditional classification accuracy measure is inappropriate in this context, we developed an evaluation method for set-valued classification based on the E-Measure. Furthermore, we enhanced our algorithm by not relying on the typical support/confidence framework, and instead mining for the best possible rules above a user-defined minimum confidence and within a desired range for the number of rules. This avoids long mining times that might produce large collections of rules with low predictive power. For this purpose, we developed a heuristic function to determine an initial minimum support and then adjusted it using a binary search strategy until a number of rules within the given range was obtained. We implemented all of our techniques described above in WEKA, an open source suite of machine learning algorithms. We used several datasets from the UCI Machine Learning Repository to test and evaluate our techniques.

Search and summarization of streaming social media, such as Twitter, requires the ongoing analysis of large volumes of data with dynamically changing characteristics. Tweets are short and repetitious -- lacking context and structure -- making it difficult to generate a coherent synopsis of events within a given time period. Although some established algorithms for frequent itemset analysis might provide an efficient foundation for synopsis generation, the unmodified application of standard methods produces a complex mass of rules, dominated by common language constructs and many trivial variations on topically related results. Moreover, these results are not necessarily specific to events within the time period of interest. To address these problems, we build upon the Linear time Closed itemset Mining (LCM) algorithm, which is particularly suited to the large and sparse vocabulary of tweets. LCM generates only closed itemsets, providing an immediate reduction in the number of trivial results. To reduce the impact of function words and common language constructs, we apply a filltering step that preserves these terms only when they may form part of a relevant collocation. To further reduce trivial results, we propose a novel strengthening of the closure condition of LCM to retain only those results that exceed a threshold of distinctiveness. Finally, we perform temporal ranking, based on information gain, to identify results that are particularly relevant to the time period of interest. We evaluate our work over a collection of tweets gathered in late 2012, exploring the efficiency and filtering characteristic of each processing step, both individually and collectively. Based on our experience, the resulting synopses from various time periods provide understandable and meaningful pictures of events within those periods, with potential application to tasks such as temporal summarization and query expansion for search.

Search and summarization of streaming social media, such as Twitter, requires the ongoing analysis of large volumes of data with dynamically changing characteristics. Tweets are short and repetitious -- lacking context and structure -- making it difficult to generate a coherent synopsis of events within a given time period. Although some established algorithms for frequent itemset analysis might provide an efficient foundation for synopsis generation, the unmodified application of standard methods produces a complex mass of rules, dominated by common language constructs and many trivial variations on topically related results. Moreover, these results are not necessarily specific to events within the time period of interest. To address these problems, we build upon the Linear time Closed itemset Mining (LCM) algorithm, which is particularly suited to the large and sparse vocabulary of tweets. LCM generates only closed itemsets, providing an immediate reduction in the number of trivial results. To reduce the impact of function words and common language constructs, we apply a filltering step that preserves these terms only when they may form part of a relevant collocation. To further reduce trivial results, we propose a novel strengthening of the closure condition of LCM to retain only those results that exceed a threshold of distinctiveness. Finally, we perform temporal ranking, based on information gain, to identify results that are particularly relevant to the time period of interest. We evaluate our work over a collection of tweets gathered in late 2012, exploring the efficiency and filtering characteristic of each processing step, both individually and collectively. Based on our experience, the resulting synopses from various time periods provide understandable and meaningful pictures of events within those periods, with potential application to tasks such as temporal summarization and query expansion for search.