| Modern web search engines are so powerful that thousands to millions of related search results can be retrieved for a common query. Existing web search interfaces will display representative information, such as the result documents' titles/snippets and result page screenshots, in a one-dimensional ranked list. Consequently, the number of documents that can be effectively presented on one screen is usually limited to between 5 and 10. Since most users tend to browse only the first few pages regardless of whether they can find desired information or not, users fail to fully benefit from all the relevant results found by the search engines.;In this dissertation, we proposed a visualized search result displaying method based on co-clustering and key phrase extraction that expresses search results from three different perspectives. By including text, classification and clustering information in the display, the resulting system substantially improves the information compactness of the results displayed. In addition, it provides a few unique post-search interaction methods that enable a user to effectively browse and manage a large number of results conveniently.;Co-clustering algorithms now available are of either single-level or binary. We designed a non-binary co-clustering method that can organize search results into multi-level groups. Existing key phrase extraction approaches could not remove noisy phrases due to the inability of phrase relationships structures used to capture phrase independence effectively. We proposed a novel data structure for maintaining phrase relationships and a key phrase extraction method that is more complete in removing dependent phrases.;Our experiments demonstrate that our proposed interface can help users to find desired search results more quickly for implicit queries and that the proposed key phrase extraction algorithm can achieve both better effectiveness and efficiency than two existing state-of-the-art methods. |
PDF Full Text Request