Catálogo de publicaciones - libros

Data mining is a newly developed and emerging area of computational intelligence that offers new theories, techniques, and tools for analysis of large data sets. It is expected to offer more and more support to modern organizations which face a serious challenge of how to make decision from massively increased information so that they can better understand their markets, customers, suppliers, operations and internal business processes. This paper discusses fuzzy decision-making using the Grey Related Analysis method. Fuzzy models are expected to better reflect decision maker uncertainty, at some cost in accuracy relative to crisp models. Monte Carlo Simulation, a data mining technique, is used to measure the impact of fuzzy models relative to crisp models. Fuzzy models were found to provide fits as good as crisp models in the data analyzed.

There have been numerous and various complex networks with the development of science, technology and human society, such as the Internet, the World Wide Web, the network of air lines, large-scale electric power networks, the structure of a piece of Very Large-Scale Integration (VLSI),the human social relationships, the neural networks, and the spreading path net of an infectious disease, etc. Even in the study on semantics of human language, the relationship between synonyms can also be represented and analyzed via complex networks. Most researchers widely apply the parameters of degree distribution, clustering coefficient and average distance to analyze efficiently the uncertainty of complex networks.

Existing association rule mining algorithms are specifically designed to find strong patterns that have high predictive accuracy or correlation. Many useful patterns, for example, out-expectation patterns with low supports, are certainly pruned for efficiency in these mining algorithms. This talk introduces our ongoing research developing novel theories, techniques and methodologies for discovering hidden interactions within data, such as class-bridge rules and out-expectation patterns. These patterns are essentially different from traditional association rules, but are much more useful than traditional ones to applications such as cross-sales, trend prediction, detecting behavior changes, and recognizing rare but significant events. This delivers a paradigm shift from existing data mining techniques. In addition, the system of applying these techniques to stock market is briefly presented.

In many languages, the English word “computer” is often literally translated to “the counting machine.” Counting is apparently the most elementary operation that a computer can do, and thus it should be trivial to a computer to count. This, however, is a misconception. The apparently simple operation of enumeration and counting is actually computationally hard. It is also one of the most important elementary operation for many data mining tasks. We show how capital counting is for a variety of data mining applications and how this complex task can be achieved with acceptable efficiency.

Most existing text classification methods (and text mining methods at large) are based on representing the documents using the traditional vector space model. We argue that important information, such as the relationship among words, is lost. We propose a term graph model to represent not only the content of a document but also the relationship among the keywords. We demonstrate that the new model enables us to define new similarity functions, such as considering rank correlation based on PageRank-style algorithms, for the classification purpose. Our preliminary results show promising results of our new model.

Web transaction data between web visitors and web functionalities usually convey users’ task-oriented behavior patterns. Clustering web transactions, thus, may capture such informative knowledge, in turn, build user profiles, which are associated with different navigational patterns. For some advanced web applications, such as web recommendation or personalization, the aforementioned work is crucial to make web users get their preferred information accurately. On the other hand, the conventional web usage mining techniques for clustering web objects often perform clustering on usage data directly rather than take the underlying semantic relationships among the web objects into account. (LSA) model is a commonly used approach for capturing semantic associations among co-occurrence observations.. In this paper, we propose a LSA-based approach for such purpose. We demonstrated usability and scalability of the proposed approach through performing experiments on two real world datasets. The experimental results have validated the method’s effectiveness in comparison with some previous studies.

Mining association rules is an important problem in data mining. Algorithms for mining boolean data have been well studied and documented, but they cannot deal with quantitative and categorical data directly. For quantitative attributes, the general idea is partitioning the domain of a quantitative attribute into intervals, and applying boolean algorithms to the intervals. But, there is a conflict between the minimum support problem and the minimum confidence problem, while existing partitioning methods cannot avoid the conflict. Moreover, we expect the intervals to be meaningful. Clustering in data mining is a discovery process which groups a set of data such that the intracluster similarity is maximized and the intercluster similarity is minimized. The discovered clusters are used to explain the characteristics of the data distribution. The present paper will propose a novel method to find quantitative association rules by clustering the transactions of a database into clusters and projecting the clusters into the domains of the quantitative attributes to form meaningful intervals which may be overlapped. Experimental results show that our approach can efficiently find quantitative association rules, and can find important association rules which may be missed by the previous algorithms.

Mining association rules and correlation relationships have been studied in the data mining field for many years. However, the rules mined only indicate association relationships among variables in an interested system. They do not specify the essential underlying mechanism of the system that describe causal relationships. In this paper, we present an approach for mining causal relationships among attributes and propose a potential application in the field of bioinformatics. Based on the theory of causal diagram, we show the properties of our approach.

Existing mining association rules in relational tables only focus on discovering the relationship among large data items in a database. However, association rule for significant rare items that appear infrequently in a database but are highly related with other items is yet to be discovered. In this paper, we propose an algorithm called Extraction Least Pattern (ELP) algorithm that using a couple of predefined minimum support thresholds. Results from the implementation reveal that the algorithm is capable of mining rare item in multi relational tables.

Efficient discovery of frequent patterns from large databases is an active research area in data mining with broad applications in industry and deep implications in many areas of data mining. Although many efficient frequent-pattern mining techniques have been developed in the last decade, most of them assume relatively small databases, leaving extremely large but realistic datasets out of reach. A practical and appealing direction is to mine for closed itemsets. These are subsets of all frequent patterns but good representatives since they eliminate what is known as redundant patterns. In this paper we introduce an algorithm to discover closed frequent patterns efficiently in extremely large datasets. Our implementation shows that our approach outperforms similar state-of-the-art algorithms when mining extremely large datasets by at least one order of magnitude in terms of both execution time and memory usage.