Improving web retrieval precision based on semantic relationships and proximity of query keywords

Abstract

Based on recent studies, the most common queries in Web searches involve one or two keywords. While most Web search engines perform very well for a single-keyword query, their precisions is not as good for queries involving two or more keywords. Search results often contain a large number of pages that are only weakly relevant to either of the keywords. One solution is to focus on the proximity of keywords in the search results. Filtering keywords by semantic relationships could also be used. We developed a method to improve the precison of Web retrieval based on the semantic relationships between and proximity of keywords for two-keyword queries. We have implemented a system that re-ranks Web search results based on three measures: first-appearance term distance, minimum term distance, and local appearance density. Furthermore, the system enables the user to assign weights to the new rank and original ranks so that the result can be presented in order of the combined rank. We built a prototype user interface in which the user can dynamically change the weights on two different ranks. The result of the experiment showed that our method improves the precision of Web search results for two-keyword queries. © Springer-Verlag Berlin Heidelberg 2006.