Catálogo de publicaciones - libros

In this paper, we propose the effective similarity measures on which the similarity-based pattern retrieval is based. Both data sequences and query sequences are partitioned into segments, and the query processing is based upon the comparison of the features between data and query segments, instead of scanning all data elements of entire sequences. We conduct experiments on multidimensional data sequences that are generated by extracting features from video streams, and show the effectiveness of the proposed measures.

In this paper, we apply data mining techniques to construct intrusion detection patterns. We mine both system audit data and network traffic data for consistent and useful patterns of program and user behavior, and use an iterative low-frequency-finder mining algorithm to find the low frequency but important patterns.

Automatic term extraction (ATR) is an important problem in natural language processing. But most of extraction methods focus on the extraction of multiword units. Inevitably, many common words (or phrases) as terms are extracted at the same time. In this paper, we propose a hybrid method for automatic extraction of term from domain-specific un-annotated Chinese documents by means of linguistics knowledge and statistical techniques, taking dual filtering strategy and introducing a weight formula to filter term candidates. The results of the research indicate that our system is more efficient and precise than previous methods.

Data integration system usually runs on unpredictable and volatile environments. Query cost model should be update with the changes of the environment. In this paper, we tackle this problem by evolving the cost model so that it can adapt to the environment change and keep up-to-date. Firstly, the factors causing the system environment to change are analyzed and different methods are proposed to deal with these changes. Then an architecture for evolving a cost model in dynamic environment is proposed. Our experimental results show the architecture of evolving a cost model in dynamic environment can well capture changes of environment and keep cost models up-to-date.

Web page classification is an important research direction of web mining. In the paper, a SVM method of web page classification is presented. It include four steps: (1) using analysis module to extract the core text and structural tags from a web page; (2) adopting the improved VSM model to generate the initial feature vectors based on the core text of web page; (3) adjusting weights of the selected features based on structural tags in web page to generate the base SVM classifier; (4) combining the base classifiers produced by iteration based on Boosting mechanism to obtain the target SVM classifier. The experiment of web page classification shows that the approach presented is efficient.

Fuzzy neural networks have been successfully applied to analyze/generate predictive rules for medical or diagnostic data. This paper presents selected membership functions extracted by a fuzzy neural network named NEWFM. The selected membership functions can capture the concentrated and essential information without sacrificing the classification capability. To verify the performance of the NEWFM, the well-known data set of Wisconsin breast cancer is performed. We applied NEWFM model to extract fuzzy membership functions for the UCI antibody deficiency syndrome diagnosis. Then selected features obtained by non-overlapped area measurement method are presented.

Large-scale genome-wide genetic profiling using markers of single nucleotide polymorphisms (SNPs) has offered the opportunities to investigate the possibility of using those biomarkers for predicting genetic risks. Because of the special data structure characterized with a high dimension, signal-to-noise ratio and correlations between genes, but with a relative small sample size, the data analysis needs special strategies. We propose a robust data reduction technique based on a hybrid between genetic algorithm and support vector machine. The major goal of this hybridization is to fully exploit their respective merits (e.g., robustness to the size of solution space and capability of handling a very large dimension of features) for identification of key SNP features for risk prediction. We have applied the approach to the Genetic Analysis Workshop 14 COGA data to predict affection status of a sib pair based on genome-wide SNP identical-by-decent (IBD) informatics. This application has demonstrated its potential to extract useful information from the massive SNP data.

Recent advances in SNPs that allow genome-wide profiling of complex biological phenotypes have offered the golden opportunities to unravel the high-order mechanisms and have also motivated development of the corresponding analysis strategies. Here, we design four novel comprehensive association criteria concerning both informatics of IBD statistic and genomic context. Application of these criteria along with sliding window and permutation test to 100 simulated replicates for two American populations to extract the relevant SNPs for alcoholism from sib-pair IBD profiles of pedigrees demonstrates that the proposed new approaches have successfully identified most of the simulated true loci, thus implicating that IBD statistic and genomic context could be used as the informatics for mining the underlying genes for complex human diseases. Compared with the classical Haseman-Elston method, our strategy is more efficient and simpler.

The paper presents a permutation-based algorithm for predicting RNA secondary structure. It is practicable, and can be used to predict real RNA molecules. The conception of permutation is introduced, which is the start point of our algorithm. Individual is represented as a permutation of stem list. Crossover operator, mutation operator, and selection strategy are designed to be compatible with such an individual representation. At the end of the paper, a comparison between our result and that from RNAstructure is outlined. It is proved that our algorithm has achieved comparable or better result than RNAstructure.

Gene expression microarray technology provides the global information on transcriptional activities of essentially all genes simultaneously, and it thus promotes the new application of traditional feature selection methods in the fields of molecular biology and life sciences. The basic strategy for the traditional feature selection methods is to seek for a single gene subset that leads to the best prediction of biological types, for example tumor versus normal tissues. Because of complexities and genetic heterogeneities of biological phenotypes (e.g. complex diseases), robust computational approaches are desirable to achieve high generalization performance with multiple classifiers and perturbations of the data structures. The purpose of this study is to develop an ensemble decision approach to analysis of multiple heterogeneous phenotypes. The results from an application to a lymphoma data of five subtypes indicate that the proposed analysis strategy is feasible and powerful to perform biological subtype.