Catálogo de publicaciones - libros

We put forward a new conception, , which can describe the regularities of changes over time in association rules. The dynamic association rule is different in that it contains not only a support and a confidence but also a and a . During the mining process, the data used for mining is divided into several parts according to certain time indicators, such as years, seasons and months, and a support vector and a confidence vector for each rule are generated which show the support and the confidence of the rule in each subsets of the data. By using the two vectors, we can not only find the information about the rules’ changes with time but also predict the tendencies of the rules, which ordinary association rules can not offer.

In the field of cluster analysis, most of existing algorithms are developed for small data sets, which cannot effectively process the large data sets encountered in data mining. Moreover, most clustering algorithms consider the contribution of each sample for classification uniformly. In fact, different samples should be of different contribution for clustering result. For this purpose, a novel typical-sample-weighted clustering algorithm is proposed for large data sets. By the atom clustering, the new algorithm extracts the typical samples to reduce the data amount. Then the extracted samples are weighted by their corresponding typicality and then clustered by the classical fuzzy -means (FCM) algorithm. Finally, the Mahalanobis distance is employed to classify each original sample into obtained clusters. It is obvious that the novel algorithm can improve the speed and robustness of the traditional FCM algorithm. The experimental results with various test data sets illustrate the effectiveness of the proposed clustering algorithm.

This paper proposes the problem of mining weighted generalized fuzzy association rules with fuzzy taxonomies (WGF-ARs). It is an extension of the generalized fuzzy association rules with fuzzy taxonomies problem. In order to reflect the importance of different items, the notion of generalized weights is introduced, and leaf-node items and ancestor items are assigned generalized weights in our WGF-ARs. The definitions of weighted support and weighted confidence of WGF-ARs is also proposed. Then a new mining algorithm for WGF-ARs is also proposed, and several optimizations have been applied to reduce the computational complexity of the algorithm.

Different from familiar clustering objects, text documents have sparse data spaces. A common way of representing a document is as a bag of its component words, but the semantic relations between words are ignored. In this paper, we propose a novel document representation approach to strengthen the discriminative feature of document objects. We replace terms of documents with concepts in WordNet and construct a model named Concept CHain Model(CCHM) for document representation. CCHM is applied in both partitioning and agglomerative clustering analysis. Hierarchical clustering processes in different levels of concept chains. The experimental evaluation on textual data sets demonstrates the validity and efficiency of CCHM. The results of experiments with concept show the superiority of our approach in hierarchical clustering.

Time series are comprehensively appeared and developed in many applications. Similarity search under time warping has attracted much interest between the time series in the large databases. DTW (Dynamic Time Warping) is a robust distance measure and is superior to Euclidean distance. Nevertheless, it is more unfortunate that DTW has a quadratic time and the false dismissals are come forth since DTW distance does not satisfy the triangular inequality. In this paper, we propose an efficient range query algorithm based on a new similarity search method under time warping. When our range query applies for this method, it can remove the significant non-qualify time series as early as possible. Hence, it speeds up the calculation time and reduces the number of scanning the time series. Guaranteeing no false dismissals the lower bounding function is advised that consistently underestimate the DTW distance and satisfy the triangular inequality. Through the experimental results, our range query algorithm outperforms the existing others.

In this paper, an efficient and robust scene change detection algorithm is proposed by using low-level audiovisual features and several classification methods. The proposed algorithm consists of three stages. The first stage is shot boundary detection by using Support Vector Machine (SVM) and the second stage is the scene boundary detection using shot clustering based on visual information. In the last stage, the scene boundary correction with audio information is described.

Based on the FP-tree data structure, this paper presents an efficient algorithm for mining the complete set of positive correlated item pairs. Our experimental results on both synthetic and real world datasets show that, the performance of our algorithm is significantly better than that of the previously developed Taper algorithm over practical ranges of correlation threshold specifications.

As a simple, effective and nonparametric classification method, kNN algorithm is widely used in text classification. However, there is an obvious problem: when the density of training data is uneven it may decrease the precision of classification if we only consider the sequence of first k nearest neighbors but do not consider the differences of distances. To solve this problem, we adopt the theory of fuzzy sets, constructing a new membership function based on document similarities. A comparison between the proposed method and other existing kNN methods is made by experiments. The experimental results show that the algorithm based on the theory of fuzzy sets (fkNN) can promote the precision and recall of text categorization to a certain degree.

Traditional information retrieval technologies can satisfy users’ needs to some extent. But they cannot satisfy any query from different backgrounds, with different intentions and at different time because of their all-purpose characteristics. An integrated searching algorithm by combining filtering with collaborative technologies is presented in this paper. The user model is represented as the probability distribution over the domain classification model. A method of computing similarity and a method of revising user model are provided. Compared with the vector space model, the probability model is more effective on describing users’ interests. Furthermore, collaborative-based technologies are used, and as a result the scalability of the new algorithm is highly enhanced.

We present a cost model for predicting the performance of R-tree and its variants. Optimization base on the cost model can be apply in R-tree construction. we construct a new R-tree variant called CR*-tree using this optimization technique. Experiments have been carried out ,results show that relative error of the cost model is around 12.6%,and the performance for querying CR*-tree has been improved 4.25% by contrast with R*-tree’s.