1
Learning-Based Concept-Hierarchy Refinement through Exploiting Topology, Content and Social Information
Tsung-Ting Kuo, Shou-De Lin MSLab, GINM, NTU
Introduction
This is the readme for the source code of the "Learning-Based Concept-Hierarchy Refinement through Exploiting Topology, Content and Social Information" research. The source code, training data and other related resources can be downloaded on our website: http://www.csie.ntu.edu.tw/~d97944007/refinement/
Software Platform
We use Java as our development platform. Please install Java 1.6.0 or above in order to execute or modify our source code.
Hardware Platform
We suggest using AMD Opteron 2350 2.0GHz Quad-core CPU or above, with 32GB RAM or above to run the program. For the InterPro dataset, only 2GB RAM is required.
Acknowledgement
We used these libraries in our program:
Apache Common Math 1.2 or above http://commons.apache.org/math
Tim’s Java Utilities http://www.csie.ntu.edu.tw/~d97944007/utility Weka 3.5.8 or above http://www.cs.waikato.ac.nz/ml/weka
Getting Started
1. Please download "Full Package" from our website (http://www.csie.ntu.edu.tw/~d97944007/refinement/) and extract it. The directories are listed as follows:
src Source code of our program
jar JAR file of our program (for quick start) lib Directory to put required libraries yahoo Data files for content features html Data files for social features
2. Please download prerequisite libraries listed in Acknowledgement section and put the JAR files in the lib directory.
2
3. Among the source codes in the src directory or JAR file in the jar directory, three Java classes are designed to be run directly:
acmtree.ACMTreeAutoRunMain Main class for the ACM CCS dataset acmtree.DOAJAutoRunMain Main class for the DOAJ dataset acmtree.InterProAutoRunMain Main class for the InterPro dataset
4. All possible concept hierarchy refinement methods are defined in the acmtree.Method class, including:
Random Similarity_Level Similarity_Sibling Similarity_Children Similarity_Frequency Similarity_Name Similarity_Page Similarity_Jaccard Similarity_NGD Similarity_Coauthor Similarity_Sequence Similarity_OneNorm Learning
5. While using Learning as method, the features for learning are defined in the acmtree.Feature class, including:
Topology Content Social All
6. While using Learning as method, except for using Topology as feature, we can set the configuration of enrichment in the acmtree.Enrichment class, including:
Enable Disable
3
7. After execution, the results are saved in the following files:
results_acmtree.csv Results on ACM CCS dataset results_doaj.csv Results on DOAJ dataset results_gpcr.csv Results on InterPro dataset
Note
1. Please remember to include all prerequisite libraries in the Java class path (ex. use -cp command line argument to run Java).
2. Please also remember to set memory-related arguments of Java Virtual Machine (ex. use -Xms512M
-Xmx30G to run Java). We suggest using 32G RAM for running the program to avoid out of memory error. If machine with such large memory is not available, please run acmtree.InterProAutoRunMain solely, because the InterPro dataset requires about 2G RAM only.
3. A command line example is shown as below. It should be noted that for some operating system, the separating symbol of the -cp argument is colon, while other is semicolon. Also, please change the value of -Xmx argument to 32G when running on the ACM CCS or DOAJ datasets. Besides, the versions of downloaded libraries might differ from those used in the following example.
java -Xms512M -Xmx2G -cp .:jar/acmtree.jar:lib/utility.jar:lib/commons-math-1.2.jar:lib/weka.jar acmtree.InterProAutoRunMain
Please let us know if you have any question or suggestion. We appreciate your time for using our source code.
Tsung-Ting Kuo, Ph.D. Candidate of MSLab, GINM, NTU [email protected] Shou-De Lin, Assistant Professor of MSLab, GINM, NTU [email protected]
