"webhooks": [
{
"url": "",
"name": "main_webhook",
"headers": [
{
"name": "Ocp-Apim-Subscription-Key",
"value": ""
},
{
"name": "Content-Type",
"value": "application/json"
}
]
}
],Chatbot Framework Implementation
Tools and Dependencies
IBM Watson Assistant is used for the chatbot framework. It's easy to implement and modify, even for non-IT people.
Bing API is a web search API for real-time conversations.
Azure Cognitive Service for Language is used to evaluate the results from Bing API and provide more reasonable answers.
Azure Functions helps transfer information seamlessly between the chatbot generation service, Watson Assistant, and other Azure services.
Implementation Overview
Real-time Information Retrieval Webhook
To allow our chatbot answer a wider variety of questions that is not predefined during the
chatbot generation phase, we create this webhook in our customised Watson Assistant for
retrieving information during a conversation. This webhook in integrated in the template
JSON in the template_json.py file.
Because Watson Assistant webhook can only send POST
requests while Bing API only allows GET
requests, we build the Azure Function bingAccessWebHook to convert the request type.
const sendGetReqToBing = async (query, key) => {
…
const res = await fetch('https://api.bing.microsoft.com/v7.0/search' + queryStr,{
method: 'GET',
…
});
…
}
module.exports = async function (context, req) {
const body = req.body
const params = {…}
const key = req.headers["ocp-apim-subscription-key"]
let data = await sendGetReqToBing(params, key)
…
}
We later upgraded this Bing API webhook to Bing API and Azure Language Service webhook,
which is the Azure Function useQnA2ImproveAnswer.
This function calls the Bing API first
and then use Azure Language Service to evaluate the search results and conclude a reasonable
answer according to the confidence score. Although eventually not used, the Azure Function
bingAccessWebHook is not deleted from the GitHub
repository because it can offer a cheaper
solution for trusts that has extremely limited budget.
…
module.exports = async function (context, req) {
…
let data = await sendGetReqToBing(params, key)
context.log(`received ${data["webPages"]["value"].length} data from bing`)
const bingResults = data["webPages"]["value"].map(d => {
return {
"name": d.name,
"url": d.url,
"snippet": d.snippet
}
})
data = data["webPages"]["value"].map(d => d["snippet"])
…
data = await sendBingResToQnA(data, body.q)
…
const topAnswer = data.answers[0]
…
if(topAnswer.confidenceScore > miniumConfidenceScore){
result = topAnswer.answer
if(topAnswer.answerSpan.confidenceScore > miniumCS4ConciseAns){
resultConcise = topAnswer.answerSpan.text
}
}
context.res = …
}
