Expanding Relevance Feedback in the Relational Model.

Published 1 January 1997

Carol Lundquist, David Holmes, David A. Grossman, Ophir Frieder, M. Catherine McCabe

Citations5

TL;DR

In TREC 6, the relevance feedback methodology was expanded to include additional term weighting methods as well as feedback term scaling, and the relevance ranking scores between documents and queries could be improved by eliminating certain highly weighted terms.

Abstract

In TREC-6, we participated in both the automatic and manual tracks for category A. For the automatic runs, we used the short versions of the queries and enhanced our existing prototype by expanding the relevance feedback methodology to include additional term weighting methods (i.e., the typical “ltc-lnc ” or “nidf ” weights) as well as feedback term scaling. We also experimented with eliminating infrequently occurring terms to determine if the relevance ranking scores between documents and queries could be improved by eliminating certain highly weighted terms. For our manual runs, we used pre-defined concept lists with terms from the concept lists combined in different ways. We continued to use the AT&amp;T DBC-1012 Model 4 parallel database machine as the platform for our information retrieval system which continues to be implemented in the relational database model using unchanged SQL.

Keywords

Computer Science

ACM SIGIR ForumPivoted Document Length Normalization

863 Citations2017Amit Singhal, Chris Buckley +1 more

Pivoted normalization is presented, a technique that can be used to modify any normalization function thereby reducing the gap between the relevance and the retrieval probabilities, and two new normalization functions--pivoted unique normalization and piuotert byte size normalization are presented.

New Retrieval Approaches Using SMART: TREC 4.

322 Citations1995Chris Buckley, Amit Singhal +2 more

The SMART information retrieval project emphasizes completely automatic approaches to the understanding and retrieval of large quantities of text in TREC 4, performing runs in the routing, ad-hoc, confused text, interactive, and foreign languange environments.

Journal of the American Society for Information ScienceIntegrating structured data and text: A relational approach

102 Citations1997David A. Grossman, Ophir Frieder +2 more

This thesis is a parallel approach that allows users to combine both structured and unstructured data in a single query that provides a less than five percent degradation over corrupted data as compared with non-corrupted data.

Information Processing & ManagementRetrieval effectiveness of proper name search methods

93 Citations1996Ulrich Pfeifer, Thomas Poersch +1 more

These measures deal with phonetic similarity, typing errors and plain string similarity and it is shown experimentally that all three approaches lead to significantly higher retrieval quality than plain identity.

Improving relevance feedback in the vector space model

24 Citations1997Carol Lundquist, David A. Grossman +1 more

Several techniques to improve relevance feedback are discussed including calibrating the number of top-ranked documents or feedback terms used for relevance feedback, clustering the top- ranked documents, changing the term weighting formula, and scaling the weight of the feedback terms.

View PDF

Using Relevance Feedback within the Relational Model for TREC-5.

11 Citations1996David A. Grossman, Carol Lundquist +4 more

For TREC-5, the existing prototype that implements relevance ranking using the AT&T DBC-1012 Model 4 parallel database machine is enhanced to include relevance feedback and relevance feedback was implemented in the special purpose IR prototype.

Text REtrieval ConferenceImproving accuracy and run-time performance for TREC-4

11 Citations1995David A. Grossman, David Holmes +3 more

For TREC-4, the existing prototype that implements relevance ranking using the AT&T DBC-1012 Model 4 parallel database machine is enhanced and query reduction based on term frequency was shown to result in dramatically improved run-time performance and, in many case, resulted in little or no degradation of precision/recall.