Systems developped for the DEFT2013 evaluation campaign
DEFT2013 evaluation campaign http://deft.limsi.fr/2013/index.php?id=1&lang=en Wikimeta Lab System
Paper is on ResearchGate at: http://goo.gl/vSVBU
For information contact: twitter @ericcharton twitter @mjmrsc
This software is free to use, modify and redistribute under Creative Commons by-nc/3.0 License Term http://creativecommons.org/licenses/by-nc-sa/3.0/
This license means, do anything with it but make money. (if you make money, share it with us :-)
(c) Dr Eric Charton & Dr Marie-Jean Meurs
!!! Caution !!!
This software (and data set) is intended for students as a way to experiment machine learning with Weka and for researcher as a reproducible experiment of DEFT 2013. THIS IS NOT SOMETHING EASY TO USE AND DEPLOY. You need machine learning, java and weka skills to use this package. We can provide a little help ... but not too much :-)
Enjoy E.
Folders description
Dont forget to add the path to your folder structure in the variables.vars class
DEFT2013/arff/ The ARFF files to reproduce experiments from GUI or code (students can use those directly in weka GUI with no further manipulations)
DEFT2013/corpus The feature lists (verbs, ngrams and so on) generated to build ARFF (you can regenerate those from software using classes of package "analyse")
DEFT2013/src The Java code; use it to make stats, generate arff, generate full experiments, test with our (or new) models. You can use serialized models from Weka gui.
DEFT2013/models Our models (Naive Bayes, LMT, SVM) generated with weka according to ARFF training files. Compliant with weka GUI and Java code
DEFT2013/run The csv and results generated by our Java application and sent to DEFT2013 organization as runs (final results of the paper)
DEFT2013/jar The version 3-7-9 of weka jar is there for your convenience (GNU General Public License) .
Try with Eclipse
- Download and install weka 3-7-9
- If you want to reproduce SVM experiment under eclipse
- Install LibBsvm from GUI
Tools/package manager => LibSVM
- Configure eclipse
To avoid classpath problems, please install from weka 3-7-9
From the Java project, under eclipse:
Make the classpath pointing on libraries:
- Right button on project, properties
- Java build path
- add external class folder : on folder /home/yourname/wekafiles/packages/LibSVM (exists if you used weka 3-7-9)
- add external jar /home/yourname/wekafiles/packages/LibSVM.jar
Dont forget to import weka.jar from 3-7-9.
Use the Weka GUI and explorer
- Please launch Weka with enough memory to test on ARFF files:
java -Xmx2048m -jar weka.jar
- You can use ARFF files from ARFF folder to make experiments
Same results as the Wikimeta Lab system for DEFT2013 will be obtained with same classifiers (see paper http://goo.gl/vSVBU) and a 5 fold experiment.