Skip to content

Generalizability Theory for Information Retrieval Evaluation

License

Notifications You must be signed in to change notification settings

julian-urbano/gt4ireval

Repository files navigation

Travis-CI Build Status License CRAN version CRAN downloads

gt4ireval

Provides tools to measure the reliability of an Information Retrieval test collection. It allows users to estimate reliability using Generalizability Theory and map those estimates onto well-known indicators such as Kendall tau correlation or sensitivity.

For a full background please refer to Julián Urbano, Mónica Marrero and Diego Martín, "On the Measurement of Test Collection Reliability", ACM SIGIR, 2013.

Installation

You may install the stable release from CRAN

install.packages("gt4ireval")

or the latest development version from GitHub

devtools::install_github("julian-urbano/gt4ireval", ref = "develop")

Usage

A full user manual in available in the package vignette.

As a very simple example, we can analyze the TREC-3 Ad hoc data:

head(adhoc3)
#     sys1   sys2   sys3   sys4   sys5   sys6   sys7 ...
# 1 0.2830 0.5163 0.4810 0.5737 0.5184 0.4945 0.5013 ...
# 2 0.0168 0.5442 0.3987 0.2964 0.6115 0.2354 0.1689 ...
# ...

We first run a G-study,

ah3.g <- gstudy(adhoc3, drop = 0.25)
ah3.g
# Summary of G-Study
# 
#                  Systems     Queries Interaction
#              ----------- ----------- -----------
# Variance       0.0028117    0.028093    0.010152
# Variance(%)       6.8482      68.425      24.727
# ---
# Mean Sq.         0.15074     0.85296    0.010152
# Sample size           30          50        1500

then a D-study,

dstudy(ah3.g, queries = c(50, 100, 150))
# Summary of D-Study
# 
# Call:
#     queries = 50 
#   stability = 0.95 
#       alpha = 0.025 
# 
# Stability:
#                                            Erho2                                   Phi
#              -----------------------------------   -----------------------------------
#      Queries    Expected       Lower       Upper      Expected       Lower       Upper
#  ----------- ----------- ----------- -----------   ----------- ----------- -----------
#           50     0.93265     0.89311     0.96287       0.78613     0.66141     0.88039 
#          100     0.96515     0.94354     0.98109       0.88026     0.79621     0.93639 
#          150     0.97649     0.96164     0.98731       0.91686     0.85423     0.95668 
# 
# Required number of queries:
#                                            Erho2                                   Phi
#              -----------------------------------   -----------------------------------
#    Stability    Expected       Lower       Upper      Expected       Lower       Upper
#  ----------- ----------- ----------- -----------   ----------- ----------- -----------
#         0.95          69          37         114           259         130         487

and possibly map onto AP correlation, for instance,

gt2tauAP(Erho2 = c(0.93, 0.95, 0.98))
# [1] 0.7487836 0.8150692 0.9226192

License and Citation

gt4ireval is released under the terms of the MIT License.

If you use this code in your work, please cite the following paper:

@inproceedings{Urbano2013measurement,
  author = {Urbano, Juli\'{a}n and Marrero, M\'{o}nica and Mart\'{\i}n, Diego},
  booktitle = {International ACM SIGIR Conference on Research and Development in Information Retrieval},
  pages = {393--402},
  title = {{On the Measurement of Test Collection Reliability}},
  year = {2013}
}

Acknowledgements

This work is supported by an A4U postdoctoral grant and a Juan de la Cierva postdoctoral fellowship.

About

Generalizability Theory for Information Retrieval Evaluation

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages