We all come across chatbots when visiting various sites, while some of them operate behind real-human interaction, others are powered by AI.
In this article, we'll walk through building a simple AI-powered chatbot using TensorFlow and JavaScript. The chatbot will recognize user commands and respond with predefined answers.
Step-by-Step Guide
Setting Up Our Project
First, we create a new directory for our project and initialize it with npm, ensure you have Node.js installed on your system before starting this step.
mkdir chatbot cd chatbot npm init -y
Install necessary packages
We would be using the following npm packages for our simple project:
@tensorflow/tfjs
: TensorFlow.js library for machine learning.@tensorflow-models/universal-sentence-encoder
: Pre-trained Universal Sentence Encoder model for intent recognition.npm install @tensorflow/tfjs @tensorflow-models/universal-sentence-encoder
Create intents
Create a file named
intents.js
to store intents/commands. These are categories of user inputs that the chatbot will recognize (e.g., greetings, product inquiries, order status).// intents.js const intents = { greeting: ["hello", "hi", "hey", "good morning", "good evening", "howdy"], goodbye: ["bye", "goodbye", "see you later", "farewell", "catch you later"], thanks: ["thank you", "thanks", "much appreciated", "thank you very much"], product_inquiry: ["tell me about your products", "what do you sell?", "product information", "what can I buy?", "show me your products"], order_status: ["where is my order?", "order status", "track my order", "order tracking", "order update"], shipping_info: ["shipping information", "how do you ship?", "shipping methods", "delivery options", "how long does shipping take?"], return_policy: ["return policy", "how to return?", "return process", "can I return?", "returns"], payment_methods: ["payment options", "how can I pay?", "payment methods", "available payments"], support_contact: ["contact support", "how to contact support?", "customer support contact", "support info", "customer service contact"], business_hours: ["business hours", "working hours", "when are you open?", "opening hours", "store hours"] }; module.exports = { intents }
Create responses
Create another file named
responses.js
to store predefined responses. These are the predefined responses the chatbot will give based on the recognized intent.// responses.js const responses = { greeting: "Hello! How can I help you today?", goodbye: "Goodbye! Have a great day!", thanks: "You're welcome! If you have any other questions, feel free to ask.", product_inquiry: "We offer a variety of products including electronics, books, clothing, and more. How can I assist you further?", order_status: "Please provide your order ID, and I will check the status for you.", shipping_info: "We offer various shipping methods including standard, express, and next-day delivery. Shipping times depend on the method chosen and your location.", return_policy: "Our return policy allows you to return products within 30 days of purchase. Please visit our returns page for detailed instructions.", payment_methods: "We accept multiple payment methods including credit/debit cards, PayPal, and bank transfers. Please choose the method that suits you best at checkout.", support_contact: "You can contact our support team via email at support@example.com or call us at 1-800-123-4567.", business_hours: "Our business hours are Monday to Friday, 9 AM to 5 PM. We are closed on weekends and public holidays." }; module.exports = { responses };
Loading TensorFlow and the Sentence Encoder
Create a main script file named
chatbot.js
and load the necessary libraries and models, we load the universal sentence encoder model asynchronously and start the chatbot once the model is loaded.// chatbot.js const tf = require('@tensorflow/tfjs'); const use = require('@tensorflow-models/universal-sentence-encoder'); const { intents } = require('./intents'); const { responses } = require('./responses'); const readline = require('readline'); // Load the Universal Sentence Encoder model let model; use.load().then((loadedModel) => { model = loadedModel; console.log("Model loaded"); startChatbot(); });
Implementing Intent Recognition
Add a function to recognize the intent of the user's input, we embed the user input into a high-dimensional vector using the universal encoder and then track the highest similarity score based on the intent.
async function recognizeIntent(userInput) { const userInputEmb = await model.embed([userInput]); let maxScore = -1; let recognizedIntent = null; for (const [intent, examples] of Object.entries(intents)) { // Embedding the example phrases for each intent & Calculating similarity scores between the user input embedding and the example embeddings const examplesEmb = await model.embed(examples); const scores = await tf.matMul(userInputEmb, examplesEmb, false, true).data(); const maxExampleScore = Math.max(...scores); if (maxExampleScore > maxScore) { maxScore = maxExampleScore; recognizedIntent = intent; } } return recognizedIntent; }
Generating Responses
Add a function to generate responses based on the recognized intent:
async function generateResponse(userInput) { const intent = await recognizeIntent(userInput); if (intent && responses[intent]) { return responses[intent]; } else { return "I'm sorry, I don't understand that. Can you please rephrase?"; } }
Implementing Chatbot Interaction
Finally, implement the interaction loop with the chatbot by setting up the interface for reading user input from the command line, prompting the user for input and generating responses accordingly:
function startChatbot() { const rl = readline.createInterface({ input: process.stdin, output: process.stdout }); console.log("Welcome to the customer service chatbot! Type 'quit' to exit."); rl.prompt(); rl.on('line', async (line) => { const userInput = line.trim(); if (userInput.toLowerCase() === 'quit') { console.log("Chatbot: Goodbye!"); rl.close(); return; } const response = await generateResponse(userInput); console.log(`Chatbot: ${response}`); rl.prompt(); }); }
Here is the completed code for
chatbot.js
:// chatbot.js const tf = require('@tensorflow/tfjs'); const use = require('@tensorflow-models/universal-sentence-encoder'); const { intents } = require('./intents'); const { responses } = require('./responses'); const readline = require('readline'); // Load the Universal Sentence Encoder model let model; use.load().then((loadedModel) => { model = loadedModel; console.log("Model loaded"); startChatbot(); }); async function recognizeIntent(userInput) { const userInputEmb = await model.embed([userInput]); let maxScore = -1; let recognizedIntent = null; for (const [intent, examples] of Object.entries(intents)) { const examplesEmb = await model.embed(examples); const scores = await tf.matMul(userInputEmb, examplesEmb, false, true).data(); const maxExampleScore = Math.max(...scores); if (maxExampleScore > maxScore) { maxScore = maxExampleScore; recognizedIntent = intent; } } return recognizedIntent; } async function generateResponse(userInput) { const intent = await recognizeIntent(userInput); if (intent && responses[intent]) { return responses[intent]; } else { return "I'm sorry, I don't understand that. Can you please rephrase?"; } } function startChatbot() { const rl = readline.createInterface({ input: process.stdin, output: process.stdout }); console.log("Welcome to the customer service chatbot! Type 'quit' to exit."); rl.prompt(); rl.on('line', async (line) => { const userInput = line.trim(); if (userInput.toLowerCase() === 'quit') { console.log("Chatbot: Goodbye!"); rl.close(); return; } const response = await generateResponse(userInput); console.log(`Chatbot: ${response}`); rl.prompt(); }); }
To run the chatbot, execute the
chatbot.js
file:node chatbot.js
Voila! Our command output should have the chatbot running:
Conclusion
In this article, we've built a simple customer service chatbot using TensorFlow and JavaScript. The model used in this article is basic and doesn't have the ability to generate complex responses, in a real-world scenario an extensive dataset would be used to handle generating user queries.
Likewise, you can expand this project by integrating APIs using AXIOS, adding more intents and responses, or deploying it on a web platform.
Happy coding!
๐๐พ Learn more about me
๐๐พ Connect on LinkedIn
๐๐พ Subscribe to my blog