kinesis-ner

This is a exmaple to build a Named Entity Recognizer(NER) pipeline that:

1. fetch data from Amazon Kinesis stream
2. process text fields by Stanford CoreNLP and extract entities
3. store result to dynamoDB table

Getting started:

1. To build, run:

mvn package

2. Maven will generate a zip file kinesis-ner-{version}-SNAPSHOT-package.zip in the target/ folder. The zip file contains all dependencies except Stanford NLP model jar since it is too large.

(Note: To include the NLP models in package, you can remove the provided scope for coreNLP models in pom.xml before you run mvn package. Then you can skip step 3.)

3. Downlaod CoreNLP model.

4. Unzip the package in step 2.

5. Make sure your machine have permission to create/read/write Kinesis streams and DynanoDB tables.

6. Make sure all jars are in your classpath and run:

java -Xmx1536m -cp {class_path} com.chyikwei.app.KinesisNerApplication

(Note: the process will use ~1GB ram)

6. Put some data into the stream. the sample format is json with uuid, title, text fields. Example:

{
  "uuid": "04947df8-0e9e-4471-a2f9-9af509fb5801",
  "title": "news title",
  "text": "news text"
}

7. Check entities extracted from coreNLP. they will be stored in DynamoDB's ddb-news-entities table.

8. clean up AWS resources (kinesis stream, dynamoDB tables) after test. (The settings for stream & table names are here)