😺 Catsear

The Catsear Triple Scorer is a competitor in the Triple Scoring Challenge at WSDM 2017. It is based on a hybrid approach involving several modules. The answers from all modules are finally combined by a Linear Regression classifier.

• Project Setup
• Eclipse
• Maven
• Executing

Project Setup

Before starting, it is necessary to setup the Maven in your machine. Please read the documentation about how to install Maven: http://maven.apache.org/install.html

The Python-based module requires the gensim package:

pip install gensim

Then, clone the project:

git clone https://github.com/tira-io/catsear.git

[Optional] Compile main module using Eclipse

Execute the following command line to create the Eclipse project.

mvn eclipse:eclipse

Eclipse should automatically import the external libraries from the libs/ folder. Export a self-containing JAR as starpath.v0.0.1-beta.jar and save it in the main directory.

Compile main module using Maven

Before compiling the project using Maven, it is necessary to install locally the following libraries found in the libs/ directory:

• kbox-v0.0.1-alpha2.jar
• dbtrends.scc-v0.1.3-beta.jar
• dbtrends.core-v0.1.3-beta.jar
• starpath.benchmark.jar
• starpath.indexbuilder.jar
• starpath.indexbuilder.xingu.jar

Step 1

To setup the libraries above, execute the following command line for each of them:

mvn install:install-file -Dfile=<path-to-file> -DgroupId=<group-id> -DartifactId=<artifact-id> -Dversion=<version> -Dpackaging=<packaging>

You can also copy-paste the following commands:

mvn install:install-file -Dfile=libs/kbox-v0.0.1-alpha2.jar -DgroupId=org.aksw.kbox -DartifactId=kbox.kibe -Dversion=v0.0.1-alpha2 -Dpackaging=jar
mvn install:install-file -Dfile=libs/dbtrends.core-v0.1.3-beta.jar -DgroupId=org.aksw.dbtrends -DartifactId=dbtrends.core -Dversion=v0.1.3-beta -Dpackaging=jar
mvn install:install-file -Dfile=libs/dbtrends.scc-v0.1.3-beta.jar -DgroupId=org.aksw.dbtrends -DartifactId=dbtrends.scc -Dversion=v0.1.3-beta -Dpackaging=jar
mvn install:install-file -Dfile=libs/starpath.benchmark.jar -DgroupId=org.aksw.starpath -DartifactId=starpath.benchmark -Dversion=v0.0.1-beta -Dpackaging=jar
mvn install:install-file -Dfile=libs/starpath.indexbuilder.jar -DgroupId=org.aksw.starpath -DartifactId=starpath.indexbuilder -Dversion=v0.0.1-beta -Dpackaging=jar
mvn install:install-file -Dfile=libs/starpath.indexbuilder.xingu.jar -DgroupId=org.aksw.starpath -DartifactId=starpath.indexbuilder.xingu -Dversion=v0.0.1-beta -Dpackaging=jar

Step 2

Now, you must setup your pom file. Note that since the libraries are installed, you can comment/remove the <scope> and <systemPath> tags from your pom file for all the six files in the libs directory.

	<dependency>
        <groupId>org.aksw.starpath</groupId>
        <artifactId>starpah.indexbuilder.xingu</artifactId>
        <version>v0.0.1-beta</version>
        <!--scope>system</scope>                                                            #remove/comment this line
        <systemPath>${project.basedir}/libs/starpath.indexbuilder.xingu.jar</systemPath-->  #remove/comment this line
	</dependency>

Step 3

Execute MVN compile in a single jar.

mvn clean compile assembly:single

Step 4

Copy the generated jar at target/ to the root directory and rename it to starpath.v0.0.1-beta.jar. This way, it can be found by the script that launches Catsear.

mv target/WSDMTriplescoreChallenge-0.0.1-beta-jar-with-dependencies.jar starpath.v0.0.1-beta.jar

Compile other modules

Execute the script setup.sh. This script will fetch the needed files and compile the graph-cross module.

Learning

The Python-based module is based on Word2Vec. The wiki-sentences file is first pre-processed with:

python python/skipgram/process_corpus.py /path/to/wiki-sentences > processed-sentences.txt

The embeddings are learned with:

python python/skipgram/embed.py processed-sentences.txt python/skipgram/instance-sentences.bin <size>

Set 10 for the size parameter.

The Linear Regression classifier can be found in file python/super/wekastrategy.py. The used formula was found by maximizing the accuracy on the training set using 10-fold cross-validation. To reproduce the experiments, download Weka and create the CSV file to use as training set with:

python python/super/wekamake.py /path/to/input/dataset predictions-1.txt predictions-2.txt predictions-3.txt predictions-4.txt predictions-5.txt training.csv

Note that outputs predictions-*.txt from all 5 modules must have been generated before.

Prediction

The project can be easily executed by the script catsear.sh as following:

./catsear.sh -i /path/to/input/dataset [-i /path/to/another/input/dataset] -o /path/to/output/

Citing this work

TBA