This is the machine learning method implemented for the article:
Dalip, Daniel Hasan, Gonçalves, Marcos A., Cristo, Marco, & Calado, Pavel (2017). A general multiview framework for assessing the quality of collaboratively created content on web 2.0. Journal of the Association for Information Science and Technology, 68(2), 286-308.
This software is compatible with Linux only.
First, prepare a training and testing dataset in libsvm the format (examples in folder toyExample):
<class> <id_feature_1>:<val_feature_1> <id_feature_2>:<val_feature_2> ... <id_feature_n>:<val_feature_n>
where <class>
is the target class <id_feature_i>
is the ith feature id (starting with 1) and <val_feature_i>
is the ith feature value. After that, configure the multiview method setting the config variables (you can change the default parameters in the configExample.cnf or configExample_l2r.cnf). Also in this file you will assign the view for each feature. For more information, see .cnf
file comments. To run this program, use:
java -jar multiview.jar <train-file> <test-file> <config-file>
Example:
java -jar multiview.jar toyExample/train_svm.txt toyExample/test_svm.txt configExample.cnf
The source code is available at the MultiviewMethod
folder. Note that you can use different methods by changing the XML MultiviewMethod/learning_methods.xml and creating their scripts. Regarding Learning to Rank method, we used the SVM-RANK library available at: https://www.cs.cornell.edu/peopl. Use the same format as SVM-RANK in case of L2R problems.
In case of having problems with the file multiview.jar
, you may want to recompile it:
ant -buildfile multiview_ant.xml
The dataset and results for assessing the quality of content regarding the Question and Answering Forums and Wikis datasets are avaliable for download in their respective links.