Classification-based Retrieval Methods to Enhance Information Discovery on the Web

The widespread adoption of the World-Wide Web (the Web) has created challenges both for society as a whole and for the technology used to build and maintain the Web. The ongoing struggle of informationretrieval systems is to wade through this vast pile of data and satisfy users by presenting them with information that most adequately it’s their needs. On a societal level, the Web is expanding faster than we can comprehend its implications or develop rules for its use. The ubiquitous use of the Web has raisedimportant social concerns in the areas of privacy, censorship, and access to information. On a technical level, the novelty of the Web and the pace of its growth have created challenges not only in thedevelopment of new applications that realize the power of the Web, but also in the technology needed to scale applications to accommodate the resulting large data sets and heavy loads. This thesis presents searching algorithms and hierarchical classification techniques for increasing a search service'sunderstanding of web queries. Existing search services rely solely on a query's occurrence in the document collection to locate relevant documents. They typically do not perform any task or topic-basedanalysis of queries using other available resources, and do not leverage changes in user query patterns over time. Provided within are a set of techniques and metrics for performing temporal analysis on query logs. Our log analyses are shown to be reasonable and informative, and can be used to detect changing trends and patterns in the query stream, thus providing valuable data to a search service.