NL-Service

This service computes the semantic similarity between based on a specified word2vec model. A ready to use docker image can be found here.

As default, the NL-Service is configured to listen to ip 0.0.0.0 at port 1234. This means it exposes port 1234 on all IPv4 addresses on the local machine, i.e. any computer in the network is able to access the service through the hosts ethernet ip.
If you want to use the service exclusively on your local machine (and deploy it via docker) you should change this configuration to 127.0.0.1 as described below.

Docker

Make sure you have configured docker to use more than 4GB RAM, otherwise the service will not start properly and the process may return 'killed' after a few minutes.

To be able to access the service outside of its docker container you should either

1. expose the port 1234 to the host computer
   expose the service to localhost only:

   	docker run -it --rm -p 127.0.0.1:1234:1234 jh00/nlsservice
   

   expose the service to all IPv4 addresses on the local machine (this allows e.g. other machines in the hosts network to access the service via the hosts ip address):

   docker run -it --rm -p 1234:1234 jh00/nlsservice
   

   OR
2. create a docker subnetwork, then other containers within the same docker network can access the service via its assigned ip

   docker network create --subnet 172.16.1.0/24 nl-network
   docker run -it --rm --ip 172.16.1.254 --net=nl-network jh00/nlsservice
   

Note: Starting the service takes very much time (5-10 minutes and longer depending on the computers performance).
Once the service is started it will print a message

INFO de.unisaarland.st.webtest.server.LanguageServiceServer - Server online at http://0.0.0.0:1234

You can test that the service works correctly with a simple curl command which should return "1.0" (depending on which network option you chose)

1. from within another container

   curl -X GET 'host.docker.internal:1234/word2vec/wordsimilarity?word1=test&word2=test'
   

   from the host mathine

   curl -X GET 'localhost:1234/word2vec/wordsimilarity?word1=test&word2=test'
   

2. from within another container in the same network (nl-network)

   curl -X GET 'http://172.16.1.254:1234/word2vec/wordsimilarity?word1=test&word2=test'
   

Compilation

This project requires Java 8 and is not compatible with newer Java versions right now.

to compile the project simply change into the projects root directory and run

mvn install

This will compile a runnable jar at $projectDir/language-analysis-app/target/language-analysis-app-1.0-SNAPSHOT-standalone.jar

Start the service

The service takes a very long time to start and about 4GB of RAM to load the word2vec model. Therefore it is recommented to start the service only once and let it run in the background, meanwhile any (client)application code is invoked independently from this service.

The NL-Service is started by invoking the specified jar from the step above

java -jar -Xmx4G language-analysis-app-1.0-SNAPSHOT-standalone.jar

Customize Ip and location of the word2vec model

The service is accessible via port 1234, this port is hardcoded right now.

The configuration file $projectDir/language-analysis-app/src/main/resources/application.conf is parsed to determine the ip and the location of the word2vec model.

To change the ip you have to adapt the properties languageanalysis-app.interface. Right now this value is set to ip 0.0.0.0 for ease docker deployment.
You may want to change this value to 127.0.0.1 to deploy and access the service locally only. If you need the service externally (on a different pc or docker container) you should enter the ip address of the network interface used to access the host machine (e.g. VPN).

To change the used word2vec model modify the vaulue of languageanalysis-app.language.word2vec.googleVector to your location. Per default the service assumes the google new vector to be located at /home/TestData/GoogleNews-vectors-negative300.bin.gz