ReWox

NLP Engine For ChatBot

Why is it better than other tools:

• Graph Node Editor Based Bots
• Sentiment Analysis

Services we need to run:

• RabbitMQ
• RethinkDB

How to Handle Monitor:

• Will not have a continous monitoring feature.
• A process will run once on a user request to analyse the web for a brand sentiment and store the aggregated report.
• Continous monitoring of timeline will be avaliable on paid plans.

Random Ideas:

• Learn autoresponder. (Sentence similarity based active learning)
• Connect with CRM to learn the chatbot.
• Set predefined answers for particular question.
• There is should be a module for training

Algorithm for suggesting answers:

• Compare the question with the rest of the (question, answer) tuple in the dataset.
• Keep those questions whose similarity are greater than 90%
• Compare the answer of the most matching questions with the answers of the other questions.
• If the average similarity is greater than 75% than suggest an answer otherwise ignore.
• Suggest the answers of the top 3 matching questions.

Analyze:

• Define Entities of Interests (e.g Iphone)
• Automatically fetch bad reviews of the entity.
• Determine with parts of the entity people are complaining about
• Determine with parts of the entity people are liking

Chatbot Platform:

• Intents: What action should be taken in response to user's message
• Entities: Objects we want to interact with

Key Advantages:

• Self Hosted
• Option for entity based answers.
• Self learning

Database Structure:

• Entities: { entity_name : "", values: [], }

• Intent: { name: "" questions: [], answers: [ {statement: "", isEntityBased: true}, ], action: { name: "", params: [ {isRequired: true, name: "", entity_name: "", value: ""} ], } entities: [] }

• questions: { statement: "", entities: [], }

SOCKET PAYLOAD

From server/ API endpoints to client:

• In String
• {custom: {task: "", module: ""}, payload: "", status: ""}

NLU Service Protocol:

• Bot Response: method: microservice body: { task: 'BOT_RESPOND', statement: '', //String, custom: {}, //Any object, intents: [], // Array of intent objects } Response: {custom: {} //Whatever Object you posted, server: {task: 'BOT_RESPOND', success: 'false' , statement: ''} //Any Object, result: { response: "" //String, action: {} //Action Object}}

API Reference Guide:

• NLP:

  • Named Entity Recognition: url: /api/nlp/ner method: post body: { task: 'NER', statement: '', //String, custom: {}, //Any object or string } Response: {custom: {} //Whatever Object or string you posted, result: [{start: "", end: "", label: ""}]}

  • Bot Response: url: /api/bot/respond method: post body: { task: 'BOT_RESPOND', statement: '', //String, custom: {}, //Any object } Response: {custom: {} //Whatever Object you posted, result: { response: "" //String, action: {} //Action Object}}

• Intent:

  • Create Intent: url: /api/intents/save method: post body: { name: "" questions: [], answers: [ {statement: "", isEntityBased: true}, ], action: { name: "", params: [ {isRequired: true, name: "", entity_name: "", value: ""} ], } entities: [] }

  • Fetch Intents: url: /api/intents/ method: get