This page describes a number of experiments on standard test
collections using Ivory. Each experiment is associated with a
command-line invocation to perform the experiment, command-line
invocations to evaluate the results, and a JUnit test case that
performs the experiment and checks to see if the results are as
expected (in terms of standard metrics of effectiveness). The purpose
of enumerating these experiments is for others to be able to replicate
our results, thereby providing a solid foundation for building on our
work.
Here are the experiments categorized:
- Experiments on TREC Disks 4-5: baselines,
weighted sequential dependence model, latent concept expansion
- Experiments on Wt10g: baselines
- Experiments on Gov2: baselines
- Experiments on ClueWeb09,
category B: baselines, Waterloo spam scores
- Experiments from Wang et al.'s SIGIR 2010
paper: the efficient sequential dependence model (ESD) on Wt10g,
Gov2, and Clue
- Experiments from Wang et al.'s CIKM 2010
paper: temporally constrained linear models on Wt10g, Gov2, and Clue
- Experiments from Wang et al.'s SIGIR 2011
paper: cascade model on Wt10g, Gov2, and Clue
- Experiments from Ture et al.'s SIGIR 2011
paper: cross-lingual pairwise similarity comparison on
Wikipedia.
- Experiments from Asadi and Lin's IRJ 2012
paper: document vector representations for feature extraction in
multi-stage search architectures.