nlp.js

ChatAlly nlp.js Middleware is a middleware wrapper around a trained nlp.js NLP module and provides a utility function to train an NLP module.

With the advent of Large Language Models (LLMs), Natural Language Processing (NLP) has become much easier recently. However, it is still inherently complex. nlp.js is a FOSS chatbot library provided by the Axa insurance group that can be an easy entry point.

trainNlp: Create a trained NLP module

/**
 * It is best practice to provide your NLP configuration as file 'conf.json' in
 * your current working directory.
 *
 * For examples see
 * https://github.com/jesus-seijas-sp/nlpjs-examples/tree/master/01.quickstart/03.config
 *
 * @param logger [Optional] Integrate it with your ChatAlly application by
 *    passing in an application logger.
 * @param configuration [Optional] Provide a configuration object
 *    programmatically or an array of strings with path names pointing to your
 *    training corpora.
 * @returns The trained NLP module
 */
export declare function trainNlp(
  logger?: Logger,
  configuration?: Configuration | string[]
): Promise<Nlp>

nlpjsMiddleware: Create a middleware function for an NLP module

/**
 * Create a ChatAlly middleware function for an nlp.js NLP module
 *
 * It will write the answer to the response only, if it is above the configured
 * threshold. It will not end the response.
 *
 * @param nlp The trained NLP module
 * @param options Options
 * @returns a ChatAlly middleware function
 */
export function nlpjsMiddleware(
  nlp: Nlp,
  options?: {
    /**
     * [Optional] Name for the middleware [`default="nlp.js"`]
     * NOTE: Result data from the NLP process will be put into the context under
     *    this name.
     */
    name?: string
    /**
     * [Optional] Indicates, whether an answer above the threshold should end the
     * response.
     * [`default=false`]
     */
    end?: boolean
    /**
     * [Optional] Threshold for confidence score that must be reached before
     * answering to the request, must be between 0 and 1.
     * [`default=0.8`]
     */
    threshold?: number
    /**
     * [Optional] Indicate, whether NLP should only generate an answer if there
     * is none, yet
     * [`default=false`]
     */
    onlyEmptyResponse?: boolean
  }
): Middleware<unknown>