CohEEL

A library for the automatic detection and disambiguation of knowledge base entity mentions in texts.

Execution

Programs can be run via the bin/run script. All programs need a --configuration parameter, which identifies a file under src/main/resources. This file configures required properties, such as job manager, hdfs, path to certain files etc.

Spread Wikipedia data dump to HDFS

bin/spread-wikidump.sh

Run preprocessing and classification scripts

# preprocessing: extract main data like surfaces, links, redirects, language models, etc.
bin/run --configuration cluster_tenem --program extract-main

# extract probability that a surface is linked at all
bin/prepare-surface-link-probs-program.sh
bin/run --configuration cluster_tenem --program surface-link-probs

# create training data
bin/prepare-tries.sh
# .. upload tries manually to locations specified in the configuration
bin/run --configuration cluster_tenem --program training-program
# training
mvn scala:run -Dlauncher=MachineLearningTestSuite

# classification
bin/run --configuration cluster_tenem --program classification --parallelism 10

AWS EMR Setup

To setup CohEEL on Amazon Elastic MapReduce (EMR), a proper installation of the AWS Command Line Interface is required. Use aws configure to configure the local installation. Furthermore, you have to setup your EC2 key pair name [keyname], as well as the path to your private key file [pemfile]:

aws configure set emr.key_name [keyname]
aws configure set emr.key_pair_file [pemfile]

The following command starts a cluster (named "coheel") with 20 worker instances of type m1.large:

# create a new cluster
aws emr create-cluster --name "coheel" \
    --release-label emr-4.2.0 \
    --use-default-roles \
    --applications Name=Hadoop Name=Ganglia \
    --instance-count 21 \
    --instance-type m1.large \
    --configurations '[{ "Classification": "yarn-site", "Properties": { "yarn.nodemanager.resource.cpu-vcores": "1", "yarn.nodemanager.resource.memory-mb": "5120" } }]' \
    --bootstrap-action Name="installFlink",Path="s3://coheel-conf/install-flink-0.10.1.sh"

# wait until the cluster is running and get the name of the master node by executing
aws emr describe-cluster --cluster-id [ClusterId] | grep MasterPublicDnsName | cut -d\" -f4

Connect to the master node via ssh (user hadoop and identity file [pemfile]) and install some required/useful dependencies (Maven, Git, jd, tmux)

# install some dependencies (Maven, Git, jd, tmux)
sudo wget http://repos.fedorapeople.org/repos/dchen/apache-maven/epel-apache-maven.repo -O /etc/yum.repos.d/epel-apache-maven.repo && sudo sed -i s/\$releasever/6/g /etc/yum.repos.d/epel-apache-maven.repo && sudo yum install -y apache-maven
sudo wget http://stedolan.github.io/jq/download/linux64/jq -O /usr/local/sbin/jd ; sudo chmod go+x /usr/local/sbin/jd
sudo yum install tmux git

# start a Apache Flink YARN session on the EMR cluster (using 20 workers)
yarn-session.sh -n 20 -s 1 -jm 768 -tm 4096 -Dfs.overwrite-files=true -Dtaskmanager.memory.fraction=0.5

To download and setup CohEEL run:

git clone https://github.com/stratosphere/coheel.git
cd coheel
# automatically retrieve the current cluster setup
source bin/load-aws-config.sh

Run a CohEEL program as usual (see Execution section) by choosing the cluster_aws setup

bin/run --configuration cluster_aws --program [...] --parallelism 20 ; coheel_message "CohEEL job finished!"

The coheel_message method sends an AWS SNS notification w/ some details after the program was terminated.