柯佳伶Jia-Ling, Kho張崴Wei, Chang2019-09-052017-8-232019-09-052014http://etds.lib.ntnu.edu.tw/cgi-bin/gs32/gsweb.cgi?o=dstdcdr&s=id=%22GN060147005S%22.&%22.id.&http://rportal.lib.ntnu.edu.tw:80/handle/20.500.12235/106589In this thesis, we propose a method to construct a facet hierarchy to organize the web search results dynamically. The proposed method is designed by two steps. First, we extract candidate facet terms according to a knowledge base. Second, we construct a facet hierarchy according to the candidate facet terms. We design an objective function to simulate the browsing cost when a user accesses the search results by a facet hierarchy. Accordingly, two algorithms are proposed to construct a facet hierarchy to optimize the objective function. The first one is a bottom-up approaches which select the best facet terms from the lowest level iteratively. The second one is a top-down approach, which uses an entropy function to estimate the expected browsing cost to select facet terms from the top level. Both algorithms are greedy algorithms which find optimal solutions. We evaluate the proposed methods on different distributions of access probability. The experiment results show that the facet hierarchies construct by the proposed methods achieves better performance on saving 30 to 50 percent of expected browsing cost than the one of the existing method.In this thesis, we propose a method to construct a facet hierarchy to organize the web search results dynamically. The proposed method is designed by two steps. First, we extract candidate facet terms according to a knowledge base. Second, we construct a facet hierarchy according to the candidate facet terms. We design an objective function to simulate the browsing cost when a user accesses the search results by a facet hierarchy. Accordingly, two algorithms are proposed to construct a facet hierarchy to optimize the objective function. The first one is a bottom-up approaches which select the best facet terms from the lowest level iteratively. The second one is a top-down approach, which uses an entropy function to estimate the expected browsing cost to select facet terms from the top level. Both algorithms are greedy algorithms which find optimal solutions. We evaluate the proposed methods on different distributions of access probability. The experiment results show that the facet hierarchies construct by the proposed methods achieves better performance on saving 30 to 50 percent of expected browsing cost than the one of the existing method.facet hierarchybrowsing costsemanticentropyuser behaviorencodingfacet hierarchybrowsing costsemanticentropyuser behaviorencoding