Modeling and using cross-topic relationships in information searching

Abstract

The hierarchical structures of search topics, commonly defined in many search engines, such as the ACM digital library, the Google search engine, and the Yahoo search engine, etc, are used to organize information based upon their cross-topic relationships - where users take advantages to follow when seeking information. Yet our studies of cross-topic relationships, by investigating the subset of ACM's and Yahoo's databases, show that relationships among search topics not only occupy in the hierarchical structures, but also exist beyond these structures. Interestingly then, those relationships seldom described in some hierarchical structures in turn may assist searching. Inspired by these findings, a model encompassing search topics is developed and is called Search Topic Network, ST Net, which can be applied to a wide range of search applications. Is-child and is-neighbor relations, a main constituent of the search topic network, connect related search topics together. They at the same time portray different important roles when it comes to searching; the is-child relation helps those searching with only general concepts, whereas the is-neighbor relation provides fresh information enhancing serendipitous searches. To study features brought by the search topic network, and, more importantly, to demonstrate the adaptability of the search topic network to different applications, we have therefore applied them to the incremental relevance feedback, considering the accessibility of information, and the meta-search engine, focusing on information coverage. Experiments show that the search topic network does gain improvements in these applications, thereby illustrating cross-topic relationships are useful for searching.