JarCastingchatbot-cli
Why
This is designed specifically for enterprise chat bot command line interface (CLI). By targeting enterprise chat bot, I mean:
- each chat bot has finite list of capabilities,
- users of chatbots have been communicated with its capabilities,
- developers behind chatbot has domain knowledge while building chatbot
These characteristics adding up, means it's very different from public facing general purpose chat bot. And we need a special designed CLI to help enterprise chatbot build.
What
In the heart of chatbot-cli, it contains a Parser parses users input to command with parameters. And from there, enterprise bot developers can implement different functions.
Here are parser features implemented/ planned:
- Command is defined by template with parameter placeholders.
- One command can have multiple templates defined.
- In a single template, it may have optional/ alternative formates of certain part of the command.
- Multiple commands can be matched for a certain input
- All these matched commands are rated with rank, and highest rank chose by default, but dewveloper of chatbot can choose alternatives if needed.
- Bot developer can extend the resolve logic to customize the matching logic.
To be added
- Command ranking is resolved by predefined rules together with Reinforcement Machine Learning
- Selected command and user feedback can be recorded and as factors for resolving command in future.
- User feedback cna be positive and negative, or no feedback.
How
A simple ChatBot command line parser, trained with templates.
- It's designed for Command Line Interface (CLI) for ChatBot, with predefined list of command syntax.
- This is NOT a NLP (Natural Language Processing) parser.
- It's smart and predictable.
- To Be Added Reinforcement learning
Build
public static void prepareParser() throws URISyntaxException {
URL csUrl = ClassLoader.getSystemResource("input/command-data.properties");
Path csPath = Paths.get(csUrl.toURI());
CommandSet commandSet = new FileCommandSet(csPath);
URL tsUrl = ClassLoader.getSystemResource("input/training-data.yml");
Path tsPath = Paths.get(tsUrl.toURI());
TrainingSet trainingSet = new FileTrainingSet(tsPath);
ParserBuilder parserBuilder = Parser.newBuilder();
parserBuilder.addCommandSet(commandSet);
parserBuilder.addTrainingSet(trainingSet);
parserBuilder.setTracer(System.out::println);
parser = parserBuilder.build();
}
In the above example, a Parser is created based on command data set Example.
Parse
Optional<ParseResult> optResult = parser.tryParse(command);
If optResult, is present, parse succeed. And returned ParseResult includes command.
Continuous improving
To Be Added Successfully parsing result, along with feedback, will be stored as training set, and improve parser working better next time.
Comments and suggestions
Raising issues if you have any feedback or reporting bugs.