Information access and retrieval with semantic background knowledge
Anil S. Chakravarthy · 1996
The rapid growth of on-line information on the Internet and other platforms presents a great challenge to information retrieval systems. However, improvements in information retrieval technology have not kept pace with the growth in available information. State-of-the-art information retrieval systems rely exclusively on literal matching of keywords in queries and documents. In order to shoulder a greater load in searching for information and to improve the effectiveness of searches, information retrieval systems need the ability to automatically match user queries and documents without relying on literal equivalence. This dissertation presents a framework for generalization through the incorporation of semantic knowledge into information retrieval systems. The semantic knowledge is extracted from an online Webster's dictionary and a thesaurus named WordNet. The use of semantic knowledge is guided by constructing structured representations of the queries and the text, which identify the roles being played by the salient words. Generalization is controlled by the use of application-dependent matching rules. The dissertation describes the architecture, implementation and evaluation of two applications, ImEngine and NetSerf, built using this framework. ImEngine retrieves captions of pictures and video clips using natural language queries. NetSerf is a system that enables information access by finding Internet information archives in response to user queries. The performance of ImEngine and NetSerf is compared to that of a standard keyword-based retrieval system, SMART. The dissertation also presents new techniques for the extraction of semantic knowledge from the dictionary, and for word-sense disambiguation using the dictionary and WordNet. Both ImEngine and NetSerf are evaluated with respect to the three main components of the framework: semantic knowledge, structured representations and disambiguation. (Copies available exclusively from MIT Libraries, Rm. 14-0551, Cambridge, MA 02139-4307. Ph. 617-253-5668; Fax 617-253-1690.)