Mônica Ribeiro Porto Ferreira

Data pre-processing: a new algorithm for feature selection and data discretization

Data pre-processing is a key element to improve the accuracy of data mining algorithms. In the pre-processing step, the data are treated in order to make the mining process achievable and effective. Data discretization and feature selection are two important tasks that can be performed prior to the learning phase and can significantly reduce the processing effort of the data mining algorithm. In this paper, we present Omega, a new algorithm for data discretization and feature selection. Omega performs simultaneously data discretization and feature selection. We validated Omega by comparing it with other well-known algorithms for data discretization (1R, ChiMerge and Chi2) and feature selection (DTM, Relief and Chi2). The experiments compared the effects of the pre-processing techniques in the results of the C4.5 algorithm (a well-known decision tree-based classifier). In the results, the data discretization provided by Omega generates the decision tree with one of the smallest average of the number of nodes and the feature selection given by Omega leads to one of the smallest average of error rate. These results indicates that Omega is well-suited to perform both, data discretization and feature selection, being highly appropriate for pre-processing data for data mining tasks.

An efficient framework for similarity query optimization

The increasing volume of multimedia data stored in relational database management systems (RDBMS) demands efficient ways to process similarity queries. Therefore, the query processor should provide mechanisms to express similarity queries, to interpret and translate them into equivalent expression in relational algebra, to evaluate alternative query plans and finally to execute the queries using the best plan found. In this paper, we present an effective framework to interpret, translate, select the best plan and efficiently execute similarity queries over data indexed by metric access methods. Experimental evaluation of the framework shows a reduction of up to 20% in the total time required to answer similarity queries.

Efficient Execution of Conjunctive Complex Queries on Big Multimedia Databases

This paper proposes an approach to efficiently execute conjunctive queries on big complex data together with their related conventional data. The basic idea is to horizontally fragment the database according to criteria frequently used in query predicates. The collection of fragments is indexed to efficiently find the fragment(s) whose contents satisfy some query predicate(s). The contents of each fragment are then indexed as well, to support efficient filtering of the fragment data according to other query predicate(s) conjunctively connected to the former. This strategy has been applied to a collection of more than 106 million images together with their related conventional data. Experimental results show considerable performance gain of the proposed approach for queries with conventional and similarity-based predicates, compared to the use of a unique metric index for the entire database contents.

Integrating user preference to similarity queries over medical images datasets

Large amounts of images from medical exams are being stored in databases, so developing retrieval techniques is an important research problem. Retrieval based on the image visual content is usually better than using textual descriptions, as they seldom gives every nuances that the user may be interested in. Content-based image retrieval employs the similarity among images for retrieval. However, similarity is evaluated using numeric methods, and they often orders the images by similarity in a way rather distinct from the users intention. In this paper, we propose a technique to allow expressing the users preference over attributes associated to the images, so similarity queries can be refined by preference rules. Experiments performed over a dataset with computed tomography lung images shows that correctly expressing the users preferences, the similarity query precision can increase from an average of 60% up to close to 100%, when enough interesting images exists in the database.

Combine-and-conquer: improving the diversity in similarity search through influence sampling

Result diversification methods are intended to retrieve elements similar to a given object whereas also enforcing a certain degree of diversity among them, aimed at improving the answer relevance. Most of the methods are based on optimization, but bearing NP-hard solutions. Diversity is injected into an otherwise all-too-similar result set in two phases: in the first, the search space is reduced to speed up finding the optimal solution, whereas in the second a trade-off between diversity and similarity over the reduced space is obtained. It is assumed that the first phase is achieved by applying a traditional nearest neighbor algorithm, but no previous investigation evaluated the impact of the first over the second phase. In this paper, we devised alternative techniques to execute the first phase and evaluated how obtaining a better quality set of elements in the first phase can improve the diversity. Besides the traditional nearest neighbor-based pre-selection, we also considered naive random selection, cluster-based and influence-based ones. Thereafter, extensive experiments evaluated a number of state-of-the-art diversity algorithms employed in the second phase, regarding both processing time and answer quality. The obtained results have shown that although the much more elaborated (and much more time consuming) methods indeed provide best answers, other alternatives are able to provide a better commitment regarding quality and performance. Moreover, the pre-selection techniques can reduce the total running time by up to two orders of magnitude.