Copyright (C) 2015 Naoki Otani Kyoto University
Please use the following reference for citation of work that benefits from use of any portion of the Steps GLAD software:
@inproceedings{Otani2015Quality,
author = {Otani, Naoki and Baba, Yukino and Kashima, Hisashi},
title = {Quality control for crowdsourced hierarchical classification},
booktitle = {2015 IEEE International Conference on Data Mining (ICDM)},
doi = {10.1109/ICDM.2015.83},
pages = {937--942},
year = {2015}
}
For questions and comments, plase contact Naoki Otani at: otani.naoki.65v@st.kyoto-u.ac.jp
- Install GSL, if not already installed.
- Download .tar.gz from the official page
./configure (--prefix={target dir})
make
make install
- Modify Makefile to point to the locations of the GSL if needed.
cd src && make && cd ../
.- Run the demo
./bin/steps_glad data/sample_labels.dat data/sample_hierarchy.dat
.
In the 1st row:
<#labels> <#workers> <#tasks> <#classes> <#steps>
followed by
<taskID> <workerID> <labelId>
All IDs start from 0.
See data/sample_labels.dat
.
Suppose a class hierarchy has N tiers.
<classID at N-th tier> <ClassID at 1st tier> <ClassID at 2nd tier>...<ClassID at (N-1)th tier>
All IDs start from 0 for each tier.
See data/sample_hierarchy.dat
.
./bin/steps_glad <input> <hierarchy> [options]
See ./bin/steps_glad -h
for details.
We can specify the model setting by -m <int>
option.
1
: Steps GLAD with a task dependent approach2
: Steps GLAD with a class dependent approach3
: Steps GLAD with a task-and-class dependent approach4
: Steps Rasch model (as an extention example, not presented in the paper)
mkdir sample_output
./bin/steps_glad data/sample_labels.dat data/sample_hierarchy.dat -p sample_output/model0
The results will be written in sample_output/
:
model0_alpha.csv
: worker accuracy parametersmodel0_beta.csv
: task difficulty parametersmodel0_probs.csv
: probabilities
This software was implemented based on GLAD, proposed the following paper.
Whitehill, J., Wu, T., Bergsma, J., Movellan, J. R., & Ruvolo, P. L. (2009). Whose Vote Should Count More: Optimal Integration of Labels from Labelers of Unknown Expertise. In Advances in Neural Information Processing Systems 22 (NIPS) (pp. 2035–2043).