Test Stanford CoreNLP on AWS Lambda.

Using Stanford CoreNLP system [1], please develop a simple AWS Lambda [2] function that accepts a paragraph of text in English and returns: 1) overall “sentiment” of the paragraph 2) named entities [3] recognised by the standard model. Please provide a complete repository of code with instructions for deploying the function.

[1] https://stanfordnlp.github.io/CoreNLP [2] https://aws.amazon.com/lambda/ [3] https://en.wikipedia.org/wiki/Named-entity_recognition

I have prepared a Lambda Deployment Package for your convenience. The Lambda package (jar) is available at: http://second.world.stockvirtual.info/lambda-java-example-1.0-SNAPSHOT.jar This package is ready to be deployed to Lambda.

Follow the following steps to start from the scratch:

1. create an ubuntu EC2 instance and make sure jdk and maven installed correctly.

2. create a project folder, check out this repository.

• the original CoreNLP model jar is 360 MB which is too big for Lambda. I removed some unnecessary models and installed the customized jar file to the local maven repository. The customized jar file can be found in the Lambda deployment package.

3. use the command "mvn package" to buildthe the jar file which should be located under the 'target' folder.

1. upload the package (jar file) to a S3 bucket. Because of the package size is larger than 50MB, so it needs to be uploaded to S3 before imported to Lambda.

2. create a Lambda function, set runtime to Java 8 and handling function to 'NlpTest::handleRequest'.

3. import the jar file from S3.

4. create an environment variable 'INPUT' with a paragraph of text.

5. Click on 'Save' and then click on 'Test'. The Lambda function reads in the content of the INPUT environment variable and performs ner and sentiment. The result should look like this: "Detected entity: She(PERSON) Tom(PERSON) UK(COUNTRY) George(PERSON) King(TITLE) The sentiment of the sentences are: Negative, Very positive, Negative, Neutral, Positive, Neutral, Negative, , and the overall sentiment rating is 1".