Actions-on-Google: Similar intents? - actions-on-google

Question

I'm tinkering with Actions on Google with DialogFlow + Firebase. The idea is to custom build for IoT devices not supposed by the default AoG.
In the case of Radio and TV, there are 2 intents currently:
1) Channel : It take in 3 parameters: device_name (custom entity), device_action (custom entity) and value.
Eg: please change the radio channel to 22.3
OR change the channel of the tv to 22
2) Volume : It take in 3 parameters: device_name (custom entity), device_action (custom entity) and value.
Eg: please change the radio volume to 99
OR change the vol of the tv to 22
The problem is that the agent does not seem to be able to differentiate between the two well.
Is the implementation wrong?
// Edit 9 May 2018:
index.js
// Copyright 2016, Google, Inc.
// Licensed under the Apache License, Version 2.0 (the 'License');
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an 'AS IS' BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
'use strict';
const util = require('util');
const functions = require('firebase-functions');
const {
dialogflow,
Suggestions,
BasicCard,
Button,
SimpleResponse,
} = require('actions-on-google');
const {values, concat, random, randomPop} = require('./util');
const responses = require('./responses');
/** Dialogflow Contexts {#link https://dialogflow.com/docs/contexts} */
const AppContexts = {
FACT: 'choose_fact-followup',
CATS: 'choose_cats-followup',
XXX: 'choose_xxx-followup',
};
/** Dialogflow Context Lifespans {#link https://dialogflow.com/docs/contexts#lifespan} */
const Lifespans = {
DEFAULT: 5,
};
const app = dialogflow({
debug: true,
init: () => ({
data: {
// Convert array of facts to map
facts: responses.categories.reduce((o, c) => {
o[c.category] = c.facts.slice();
return o;
}, {}),
cats: responses.cats.facts.slice(), // copy cat facts
},
}),
});
/**
* Greet the user and direct them to next turn
* #param {DialogflowConversation} conv DialogflowConversation instance
* #return {void}
*/
app.intent('Unrecognized Deep Link Fallback', (conv) => {
const response = util.format(responses.general.unhandled, conv.query);
const suggestions = responses.categories.map((c) => c.suggestion);
conv.ask(response, new Suggestions(suggestions));
});
// redirect to the intent handler for tell_fact
app.intent('choose_fact', 'tell_fact');
// Say a fact
app.intent('tell_fact', (conv, {category}) => {
const {facts, cats, xxx} = conv.data;
if (values(facts).every((c) => !c.length)) {
// If every fact category facts stored in conv.data is empty,
// close the conversation
return conv.close(responses.general.heardItAll);
}
const categoryResponse =
responses.categories.find((c) => c.category === category);
const fact = randomPop(facts[categoryResponse.category]);
if (!fact) {
const otherCategory =
responses.categories.find((other) => other !== categoryResponse);
const redirect = otherCategory.category;
const parameters = {
category: redirect,
};
// Add facts context to outgoing context list
conv.contexts.set(AppContexts.FACT, Lifespans.DEFAULT, parameters);
const response = [
util.format(responses.transitions.content.heardItAll, category, redirect),
];
// If cat facts not loaded or there still are cat facts left
if (cats.length) {
response.push(responses.transitions.content.alsoCats);
}
response.push(responses.general.wantWhat);
conv.ask(concat(...response));
conv.ask(new Suggestions(otherCategory.suggestion));
if (cats.length) {
conv.ask(new Suggestions(responses.cats.suggestion));
}
return;
}
const {factPrefix} = categoryResponse;
// conv.ask can be called multiple times to have the library construct
// a single response itself the response will get sent at the end of
// the function or if the function returns a promise, after the promise
// is resolved.
conv.ask(new SimpleResponse({
speech: concat(factPrefix, fact),
text: factPrefix,
}));
conv.ask(responses.general.nextFact);
conv.ask(new BasicCard({
title: fact,
image: random(responses.content.images),
buttons: new Button({
title: responses.general.linkOut,
url: responses.content.link,
}),
}));
console.log('hiwwxxxxxww thi is aaron');
conv.ask(responses.general.suggestions.confirmation);
});
// Redirect to the intent handler for tell_cat_fact
app.intent('choose_cats', 'tell_cat_fact');
// Say a cat fact
app.intent('tell_cat_fact', (conv) => {
const {cats} = conv.data;
console.log('this is cats data' + {cats});
const fact = randomPop(cats);
if (!fact) {
conv.contexts.delete(AppContexts.FACT);
conv.contexts.delete(AppContexts.CATS);
conv.ask(responses.transitions.cats.heardItAll);
return conv.ask(responses.general.suggestions.confirmation);
}
const {factPrefix, audio} = responses.cats;
// conv.ask can be called multiple times to have the library construct
// a single response itself. The response will get sent at the end of
// the function or if the function returns a promise, after the promise
// is resolved.
const sound = util.format(audio, random(responses.cats.sounds));
conv.ask(new SimpleResponse({
// <speak></speak> is needed here since factPrefix is a SSML string
// and contains audio.
speech: `<speak>${concat(factPrefix, sound, fact)}</speak>`,
text: factPrefix,
}));
conv.ask(responses.general.nextFact);
conv.ask(new BasicCard({
title: fact,
image: random(responses.cats.images),
buttons: new Button({
title: responses.general.linkOut,
url: responses.cats.link,
}),
}));
console.log('hiwwxxxxxww thi is aaron');
conv.ask(responses.general.suggestions.confirmation);
});
//say a tv channel
app.intent('volume', (conv, {device_name,device_action, value}) => {
var no_device_name = device_name;
var no_value = value;
var no_device_action = device_action;
var this_device_value = util.inspect(value, false, null);
var this_device_name = util.inspect(device_name, false, null);
var this_device_action = util.inspect(device_action, false, null);
console.log(this_device_action[0]);
if (no_device_name[0] == 'channel'){
console.log('inside tv but CHANNEL');
}
else{
console.log('VOLUME');
}
console.log('THIS IS VOL');
console.log(no_device_action[0]);
conv.ask(`Alright, ${device_name} VOLUM is now ${value}! `);
console.log('inside volume ' + value[0]);
console.log('inside volume' + device_name[0]);
console.log('inside volume' + device_action[0]);
console.log('hiwwxxxxxww thi is aaron');
});
//say a tv channel
app.intent('channel', (conv, {device_channel_name, device_channel_action, channel_value}) => {
console.log('THIS IS CHANNEL');
conv.ask(`Alright, ${device_channel_name} CHANNEL is now ${channel_value}! `);
console.log('inside CHANNEL ' + channel_value[0]);
console.log('inside CHANNEL' + device_channel_name[0]);
console.log('inside CHANNEL' + device_channel_action[0]);
console.log('hiwwxxxxxww thi is aaron');
});
app.intent('no_input', (conv) => {
const repromptCount = parseInt(conv.arguments.get('REPROMPT_COUNT'));
if (repromptCount === 0) {
conv.ask(`What was that?`);
} else if (repromptCount === 1) {
conv.ask(`Sorry I didn't catch that. Could you repeat yourself?`);
} else if (conv.arguments.get('IS_FINAL_REPROMPT')) {
conv.close(`Okay let's try this again later.`);
}
});
// The entry point to handle a http request
exports.factsAboutGoogle = functions.https.onRequest(app);

Answer 1

If you have a sample phrase with a part of that phrase set to represent a parameter, that part of the phrase can take on any value the parameter is defined to be. The "resolved value" column just shows what it has resolved to in your sample phrase.
So in your "channel" Intent, the phrase "tv channel to 3" can also match "radio volume to 4" since "tv" and "radio" are both of the device_name entity, and "channel" and "volume" are both of the device_action entity.
You have a few solutions:
You can turn them into one Intent that accepts all the phrases and in your webhook check the value of device_action to see what you should do.
You can keep them as separate Intents and remove the device_action parameter completely. You don't need them. Just use the synonyms in various sample phrases so the learning system gets trained about which synonyms work with which Intent.
If you are still worried about synonyms, or if it makes sense still, make them different Entity types.
For example, you may want to have "channel" and "preset" as different entities under the same Entity type. You care about the difference because it changes how you treat the number, but you're still changing the channel. "Volume" would be a different entity, since it represents something completely different.
In your comments, you asked about personalized aliases for devices. You didn't ask about more than one device type with different names (so you want to handle "tv1" and "tv2" with the same Intent, but be able to detect the difference. Both of these are handled best with option (3) above, but in slightly different ways.
If you want more than one "tv", each of those would be of the same Entity Type (for example, "device tv"), but each one would be a different entity value. It might look something like this:
"But wait!" I hear you saying "What if the user has three tvs? Or nine? Do I have to set all of these up? And what if they want to call one the 'bedroom tv' and another the 'den tv'?"
That suggests that you would keep the number of TVs and their aliases in a database of some sort (which you need to do anyway - you probably need to map their TV into some unique device ID to actually turn them on) and, when the user talks to your agent, update this with a User Entity.
User Entities have user-specific names and aliases set for just that user. You'll set them using the Dialogflow API when the user first talks to your agent. If you're using V1, you'll use the /userEntities endpoint. If you're using V2, you'll use the projects.agent.sessions.entityTypes resource.