Download SPUD for Lucene

The code for the Lucene 5.0.0 implementation of the SPUD retrieval method is available for download here. The code is free to use in which ever way you wish. The results in the table below are for default implementations of current retrieval models available in Lucene 5.0.0. Details regarding the setup for these experiments are available in this short pdf. Enjoy !

Results

Mean Average Precision (NDCG@10) of Retrieval Methods Implemented in Lucene with Default parameters

Collections	ROBUST-04	WT2G	Gov2
# of docs	~0.5M	~0.2M	25M
Topics	301-450 601-700	401-450	701-850
title only queries
BM25	0.232 (0.446)	0.242 (0.420)	0.255 (0.530)
Dir	0.241 (0.446)	0.300 (0.490)	0.284 (0.558)
Dir'	0.247 (0.455)	0.314 (0.494)	0.293 (0.563)
SPUD	0.259 (0.478)	0.315 (0.497)	0.316 (0.596)
description only queries
BM25	0.228 (0.450)	0.236 (0.437)	0.220 (0.512)
Dir	0.230 (0.429)	0.264 (0.423)	0.224 (0.491)
Dir'	0.242 (0.447)	0.260 (0.446)	0.218 (0.481)
SPUD	0.253 (0.470)	0.283 (0.475)	0.241 (0.515)
narrative only queries
BM25	0.266 (0.509)	0.266 (0.486)	0.295 (0.630)
Dir	0.251 (0.471)	0.278 (0.472)	0.274 (0.580
Dir'	0.262 (0.491)	0.271 (0.466)	0.294 (0.599)
SPUD	0.295 (0.533)	0.295 (0.503)	0.317 (0.632)