Hubot-ibmcloud-nlc NPM

hubot-ibmcloud-nlc

Adds a framework to enable Natural Language interactions in your Hubot scripts.

Getting Started

Overview

If you can't remember when the right command is exit, quit, or !q, this project can help you to easily develop a bot that can understand natural language.

You won't have to think of every possible way your users could formulate a request. With as little as 3 training statements we can recognize other statements with the same meaning. We use IBM Watson Natural Language Classifier to extract the intent of the statement.

We provide tools to help you improve your training data corpus based on what your users are actually asking.

We also help you pull parameter values from a statement entered by the user. For instance, a user says: I want to scale application xyz to 4 instances. After the natural language processing has determined that this is a scale command, the needed parameter values (appname of xyz and instances of 4) need to be pulled from the statement and forwarded to the function handling the command.

Cognitive Integration

To integrate cognitive functionality in your Hubot

Add dependencies to package.json.

dependencies: {
	"hubot-ibmcloud-cognitive-lib": ">=0.0.19"
},
devDependencies: {
	"hubot-ibmcloud-auth": ">=0.0.8",
	"hubot-ibmcloud-nlc": ">=0.0.20"
}

Include these scripts in your bot's external-scripts.json.
```
[
"hubot-ibmcloud-auth",
"hubot-ibmcloud-nlc"
]
```

Authorize your user ID to access the commands.

For more details see the documentation for hubot-ibmcloud-auth

HUBOT_IBMCLOUD_POWERUSERS=<comma-separated list of power-user emails -- no spaces!>
HUBOT_IBMCLOUD_READERUSERS=<comma-separated list of reader-user emails -- no spaces!>
HUBOT_IBMCLOUD_AUTHENTICATION_DISABLED=<only if desired, disables authentication and authorization if true)>

From Bluemix, create a Watson Natural Language Classifier service and set the following environment variables with your credentials.

HUBOT_WATSON_NLC_URL=<API URL for Watson Natural Language Classifier>
HUBOT_WATSON_NLC_USERNAME=<User Id for Watson Natural Language Classifier>
HUBOT_WATSON_NLC_PASSWORD=<Password for Watson Natural Language Classifier>
HUBOT_WATSON_NLC_CLASSIFIER=<(optional) Name for the classifier>

(Optional) Configure Cognitive Learning (feedback loop). For more information see: docs/cognitiveLearning.md
```
HUBOT_CLOUDANT_ENDPOINT=
HUBOT_CLOUDANT_KEY=
HUBOT_CLOUDANT_PASSWORD=
HUBOT_CLOUDANT_DB=nlc
```

(Optional) For Bluemix, create a Watson Alchemy service to use to find locations in statement and set the following environment variables.

HUBOT_WATSON_ALCHEMY_URL=<API URL for Watson Alchemy service>
HUBOT_WATSON_ALCHEMY_APIKEY=<apikey for specific instance of Watson Alchemy service>
HUBOT_WATSON_ALCHEMY_DATASET=<(optional) Dataset to use; default is ie-en-news>

Create a Natural Language classification / parameter definition file.

For detailed information about the contents of this file see the documentation on hubot-ibmcloud-cognitive-lib
The file should be added at <project>/src/nlc/NLC.json.
TIP: Start with just 3 unique texts per classification and add more statements only when the classifier fails to classify the statement. The training UI should help with this process.
Sample NLC.json for a weather bot.

{   
	"name": "hubot-ibmcloud-weather",
	"version": "0.0.1",
	"classes": [
		{
			"class": "weather",
			"description": "Display weather conditions",
			"emittarget": "weather.js",
			"texts": [
				"What is the weather?",
				"What are the current climate conditions?",
				"what's the weather like"
			],
			"parameters": [
				{
					"name": "location",  
					"type": "city",
					"prompt": "OK. What city would you like to know the weather for?"
				}
			]
		},
		{
			"class": "weather.precipitation",
			"description": "Display precipitation forecast",
			"emittarget": "weather.precipitation.js",
			"texts": [
				"Is it raining?",
				"Do I need to bring an umbrella?",
				"What's the chance of rain?"
			],
			"parameters": [
				{
					"name": "location",  
					"type": "city",
					"prompt": "OK. What city would you like to know the chance of rain for?"
				}
			]
		}
	]
}

Load the NLC training data into the database
- The default location for the training data is src/nlc/NLC.json
- Add the following to your package.json
```
	``` json
  	"scripts": {
  		"postinstall": "initDb path/to/NLC.json"
  	}
	```
```
- To update the database during development, remove the /databases directory and re-initialize the database.
```
	```
  	rm -rf databases
  	initDb path/to/NLC.json
	```
```
- NOTE: npm should install initDb from hubot-ibmcloud-cognitive-lib. If you get a "command not found" error, you can access this script from node_modules\.bin\initDb
Train the Natural Language classifier.
- Start your bot and say anything using natural language (ie. "I need help"). The bot should detect that a trained classifier isn't available and will start training.
- NOTE: A training session takes about 15 minutes to complete. The bot will notify you when training completes.
- NOTE: More automation is still needed to re-train the classifier. You can start a training session at any time with nlc:train.

For your scripts to handle the commands, add an emit listener. For the authorization flow to work, the emit target id in the robot.on() invocation should match the regex flow id value. These should both match the 'emittarget' value specified in the JSON file.

module.exports = (robot) => {
	robot.on('weather.precipitation', (res, parameters) => {
		robot.logger.debug(`weather.precipitation - Natural Language match.`);
		// Parameter values are obtain through the cognitive/nlc process.
		// Verify that required parameter values were succesfully obtained.
		if (parameters && parameters.location) {
			processWeather(res, paramaters.location);
		}
		else {
			robot.logger.error(`Some logged error.`);
			res.reply(i18n.__('some.user.message.with.error'));
		}
	});
	robot.respond(SHOW_WEATHER, {id: 'weather.precipitation'}, function(res) {
		robot.logger.debug(`weather.precipitation - RegEx match.`);
		// Parameter values are set via regular expression match.
		// If a required parameter value is not specified, the regex match should fail.
		processWeather(res, res.match[N]);
	});

	function processWeather(res, location){
		...
	}

Add databases to your .gitignore file.

Add test coverage.

To invoke the NLC flow, you can use robot.emit(<emittarget>). The value for emittarget is what you have used in NLC.json. Here's a sample test:

it('should start `testApp01`', function(done) {
		room.robot.on('ibmcloud.formatter', (event) => {
			expect(event.message).to.be.a('string');
			expect(event.message).to.contain('Starting application *testApp1Name* in the *testSpace* space.');
			done();
		});

		var res = { message: {text: 'I want to start app testApp1Name'}, response: room };
		room.robot.emit('bluemix.app.start', res, { appname: 'testApp1Name' });
	});

What is happening under the covers

The intended general flow for the natural language processing is as follows:

For statements not handled by the bot via regular expressions, the NLC processing within this package is invoked to find a class that best matches the statement. This is done using the Watson Natural Language classifier that was trained using the various NLC JSON definitions.
Once a class has been found for the statement, now the parameter values needed to process that command need to be pulled from the statement. The parameter definitions for the class in the NLC JSON file are used to do this:

One pass is made for each parameter to pull the parameter value from the statement using the parameter type. This is done my invoking a function for that parameter type that uses hard-coded values, the 'poc' package, and/or the Watson Alchemy service to attempt to find the right parameter value.
A second pass is made for each parameter for which a value was not obtained on the first pass. On this pass, the statement is modified to remove all parameter values successfully obtained on the first pass. The same parameter type function used in the first pass is used again on this pass.
Finally, if both passes do not provide a value for one or more required parameters, then a bot conversation is used to ask the user for each missing parameter value. When the user responds, the response is used to pull the value using the two passes above. If this still fails, the user is asked to specify just the parameter value as a single word. When the user responds to this, the response is assumed to be the parameter value.

After the class has been determined and the parameter values obtained, then the emit target associated with the class and the parameters are forwarded to the auth component to make sure that the user has the authority to issue this command.
If the user is authorized to issue the command, then the emit target associated with the command is invoked with the parameters.
The emit target listener within the command script gets control. It should validate that it has the needed parameter values and then proceed with the same command processing that occurs if it had been received via the regex path.

Commands

nlc status - Show the status of the Watson NLC instance currently being used by the bot.
nlc list|show - List all the Watson NLC instances.
nlc train|retrain - Train a new Watson NLC instance with the current training data.
nlc auto approve [on|off|true|false] - Toggle auto approve of new NLC statements learned from usage.
nlc data [className] - Show the data used to train the NLC instance.

Note: Usage of auto approve could have potential negative effects. Auto-approving the classified statements if they include keywords/entities could cause incorrect classifications for other command usages in the future.

Hubot Adapter Setup

Hubot supports a variety of adapters to connect to popular chat clients. For more feature rich experiences you can setup the following adapters:

Development

Please refer to the CONTRIBUTING.md before starting any work. Steps for running this script for development purposes:

Configuration Setup

Create config folder in root of this project.
Create env in the config folder, with the following contents:

export HUBOT_WATSON_NLC_URL=<API URL for Watson Natural Language Classifier>
export HUBOT_WATSON_NLC_USERNAME=<User Id for Watson Natural Language Classifier>
export HUBOT_WATSON_NLC_PASSWORD=<Password for Watson Natural Language Classifier>
export HUBOT_WATSON_NLC_CLASSIFIER=<(optional) Name for the classifier>

In order to view content in chat clients you will need to add hubot-ibmcloud-formatter to your external-scripts.json file. Additionally, if you want to use hubot-help to make sure your command documentation is correct. Create external-scripts.json in the root of this project, with the following contents:

[
	"hubot-help",
	"hubot-ibmcloud-formatter"
]

Lastly, run npm install to obtain all the dependent node modules.

Running Hubot with Adapters

Hubot supports a variety of adapters to connect to popular chat clients.

If you just want to use:

Terminal: run npm run start
Slack: link to setup instructions
Facebook Messenger: link to setup instructions

License

See LICENSE.txt for license information.

Contribute

Please check out our Contribution Guidelines for detailed information on how you can lend a hand.

hubot hubot-scripts cognitive nlc natural language watson natural language classifier parameter extraction ibm cloud ibm cloud

esrever

@everything-registry/sub-chunk-1875 cloudbot

8 years ago

8 years ago

8 years ago

8 years ago

8 years ago

8 years ago

8 years ago

8 years ago

8 years ago

8 years ago

8 years ago

8 years ago

8 years ago

8 years ago