IT Technical Report Series for the Year 2005

[ 2009 ( 1 ) ][ 2005 ( 3 ) ][ 2004 ( 12 ) ][ 2003 ( 9 ) ][ 2002 ( 16 ) ][ 2001 ( 12 ) ][ 2000 ( 11 ) ][ 1999 ( 15 ) ]

Technical Reports

Evolving Term-Selection Schemes for Pseudo-Relevance Feedback in Information Retrieval

Technical Report No. : NUIG-IT-201205
By : R. Cummins and C. O Riordan

Automatic query expansion in Information Retrieval aims to improve retrieval performance by overcoming the problem of term mismatch between a query and its relevant documents. Pseudo-relevance (blind) feedback techniques have been shown to be of benefit on large TREC collections in recent years. This technique analyses terms in the top few documents deemed relevant by the system, reformulates the query and runs the newly formulated query through the system again. This paper describes a method which uses Genetic Programming to evolve a scheme for selecting and weighting terms from the top ranked documents in order to expand the initial query and increase the mean average precision achieved by the system. The scheme is also used to weight the terms in the reformulated query. As a result, the genetic program has to, not only learn a scheme for identifying the best terms for expansion, but also learn a scheme which correctly weights these in relation to each other. The resulting schemes are tested on standard test collections and are shown to increase mean average precision over existing benchmark term selection schemes.

Download the Full Paper

Developing Local Term-weighting schemes in Information Retrieval using Genetic Programming

Technical Report No. : NUIG-IT-211205
By : R.Cummins, C. O Riordan

Machine Learning techniques are increasingly being applied to many areas in Information Retrieval. Genetic Programming has been shown to be a viable method for developing term-weighting schemes in IR. This paper presents term-weighting schemes that have been evolved in both a global (collection-wide) and local (within-document) context. In particular, local (within-document) weighting schemes are evolved dependent on a previously evolved global scheme and we show an increase in mean average precision over the BM25 scheme for the combined local and global scheme. An analysis of the term-frequency influence of best performing within-document scheme is shown to behave similarly to that of Okapi-tf when its term-frequency influence parameter is assigned a low value. The document normalisation part of the evolved local scheme does not perform as well as Okapi-tf on long documents. We conclude that Okapi-tf can be tuned to interact effectively with the evolved global weighting scheme presented and increase average precision over the standard BM25 scheme.

Download the Full Paper

Clinical Trials of a WLAN Patient Location System

Technical Report No. : NUIG-IT-231205
By : Aoife Deery, Desmond Chambers, Denis Moriarty, Gerard Lyons, Elizabeth Connolly

This paper describes the clinical trials run on an Indoor Positioning System that can be installed on any wireless 802.11 network. It discusses the trials, initial results collected, limitations of the system in place due to architecture and site location and the opinions of professional staff on how it could be used and adapted in the clinical setting of a hospital environment.

Download the Full Paper
 

• Disclaimer
• Privacy
• Copyright
• Accessibility