Yoo Jung An

Enriching Ontology for Deep Web Search

This paper addresses the problems of extracting instances from the Deep Web, enriching a domain specific ontology with those instances, and using this ontology to improve Web search. Extending an existing ontology with a large number of instances extracted from the Deep Web is an important process for making the ontology more usable for indexing of Deep Web sites. We demonstrate how instances extracted from the Deep Web are used to enhance a domain ontology. We show the contribution of the enriched ontology to Web search effectiveness. This is done by comparing the number of relevant Web sites returned by a search engine with a users search terms only, with the Web sites found when using additional ontology-based search terms. Experiments suggest that the ontology plus instancesapproach results in more relevant Web sites among the first 100 hits.

Assessment for Ontology-Supported Deep Web Search

Ontologies could play an important role in assisting users in their search for Web pages. This paper considers the problem of constructing domain ontologies that support users in their Web search efforts and that increase the number of relevant Web pages that are returned. To achieve this goal, this paper suggests combining Deep Web information, which consists of dynamically generated Web pages, which cannot be indexed by the existing automated Web crawlers, with ontologies. Improvements when finding deep Web sites returned by a search engine are assessed based on the framework formulated in this paper. Experimental results suggest that the proposed methods assist users in finding more relevant Web sites.

Tax Knowledge Adventure: Ontologies that Analyze Corporate Tax Transactions

Software about U.S. Income Tax has been concentrated on numerical analysis, calculating tax effects where the legal results of transactions are known. In this paper, utilizing ontologies which represent computerized knowledge, tax software can be extended to explain its computation and also educate the user about the domain, corporate tax. Ontologies can be equally useful as a tool for analyzing transactions to determine what the legal implications are. This paper includes 1) designing Corporate Tax ontology that reflects the structure and form of the U.S. Internal Revenue Code (IRC), 2) its application that supports user interactive form filling functionality. Finally, 3) it demonstrates how the end user can learn about the corporate tax consequences. The newly developed software prototype, referred to as Tax Knowledge Adventure or TKA in short, is useful not only for Tax payers but also for government officers to respond to inquiries or appeals with smart reasoning.

Minimising branch crossings in phylogenetic trees

While phylogenetic trees are widely used in bioinformatics, one of the major problems is that different dendrograms may be constructed depending on several factors. Albeit numerous quantitative measures to compare two different phylogenetic trees have been proposed, visual comparison is often necessary. Displaying a pair of alternative phylogenetic trees together by finding a proper order of taxa in the leaf level was considered earlier to give better visual insights of how two dendrograms are similar. This approach raised a problem of branch crossing. Here, a couple of efficient methods to count the number of branch crossings in the trees for a given taxa order are presented. Using the number of branch crossings as a fitness function, genetic algorithms are used to find a taxa order such that two alternative phylogenetic trees can be shown with semi-minimal number of branch crossing. A couple of methods to encode/decode a taxa order to/from a chromosome where genetic operators can be applied are also given.

Syntatic and Semantic Taxonomy of Preferential Voting Methods

Preferential voting where the voters rank candidates in order of preference plays an important role in many decision making problems and have been studied intensively. Yet there are too many variations and many popular methods are promulgated differently in different regions. Hence, some iconic conventional methods are reviewed for syntactic patterns and categorized. A nomenclature for these voting methods is suggested to reveal their syntactic patterns. Over a thousand of voting methods are devised from the conventional procedural patterns. Over 60 representative voting methods are used to reveal their semantic relationship in the form of hierarchical clustering tree. All preferential voting methods perform significantly different from the simplest plurality method.

Constructing binary decision trees for predicting Deep Venous Thrombosis

Deep Venous Thrombosis (DVT) is an intrinsic disease where blood clots form in a deep vein in the body. Since DVT has a high mortality rate, predicting it early is important. Decision trees are simple and practical prediction models but often suffer from excessive complexity and can even be incomprehensible. Here a genetic algorithm is used to construct decision trees of increased accuracy and efficiency compared to those constructed by the conventional ID3 or C4.5 decision tree building algorithms. Experimental results on two DVT datasets are presented and discussed.