Prompt user for input in Microsoft Bot Framework

In this example we are going to ask the user for name, age and a flight date. The challenge is to validate the input, store it in memory and send prompts to the user until we have all the infos. This example, which I took from the official docs (including slight changes made by myself), uses concepts from articles I posted earlier, such as Managing State.

For the age and date input, we use the Microsoft/Recognizers-Text libraries to perform the initial parsing. So we have to install the dependency first.

npm install @microsoft/recognizers-text-suite

Step 1: Creating state objects

We need to store the user’s answers in state. So let’s create conversationState and userState and pass them to our MyBot class.

// index.ts
// ...
const memoryStorage = new MemoryStorage();
const conversationState = new ConversationState(memoryStorage);
const userState = new UserState(memoryStorage);

const myBot = new MyBot(conversationState, userState);

In previous articles our bot’s “entry point” of processing the requests has been onTurn:

// index.ts
// ...
server.post('/api/messages', (req, res) => {
    adapter.processActivity(req, res, async (context) => {
        await myBot.onTurn(context);
    });
});

Now we will use run instead:

// index.ts
// ...
server.post('/api/messages', (req, res) => {
    adapter.processActivity(req, res, async (context) => {
        await myBot.run(context);
    });
});

…because this time our MyBot class extends ActivityHandler which gives us the opportunity to use on<Event> methods such as

// bot.ts
// ...
export class MyBot extends ActivityHandler {
    constructor(conversationState, userState) {
        // ...

        this.onMessage(async (turnContext, next) => {
           // ...
        });

        this.onDialog(async (context, next) => {
          // ...
        });
    }
}

Step 2: Creating state properties

  • userData: which will store name, age and date of our user
  • conversationData: which will store the last question that we asked the user
export class MyBot extends ActivityHandler {
    private userData;
    private conversationData;
    private conversationState;
    private userState;

    constructor(conversationState, userState) {
        super();
        this.conversationData = conversationState.createProperty(CONVERSATION_FLOW_PROPERTY);
        this.userData = userState.createProperty(USER_PROFILE_PROPERTY);

        // The state management objects for the conversation and user.
        this.conversationState = conversationState;
        this.userState = userState;
    }
}

Step 3: Save state changes

Next, we need to store state changes. We use onDialog for that:

constructor(conversationState, userState) {
    // ...

    this.onDialog(async (context, next) => {
        // Save any state changes. The load happened during the execution of the Dialog.
        await this.conversationState.saveChanges(context, false);
        await this.userState.saveChanges(context, false);

        // By calling next() you ensure that the next BotHandler is run.
        await next();
    });
}

Step 4: Ask the user for input

We want to react whenever the user sends a message. We add onMessage to the constructor:

constructor(conversationState, userState) {
    // ...

    this.onMessage(async (turnContext, next) => {
        const flow = await this.conversationData.get(turnContext, { lastQuestionAsked: question.none });
        const profile = await this.userData.get(turnContext, {});

        await MyBot.fillOutUserProfile(flow, profile, turnContext);

        // By calling next() you ensure that the next BotHandler is run.
        await next();
    });
}

Our method fillOutUserProfile will

  • validate the user input
  • check ConversationState for the last asked question
  • send back the next question to the user
// src/ bot.ts
static async fillOutUserProfile(flow, profile, turnContext) {
    const input = turnContext.activity.text;
    let result;
    switch (flow.lastQuestionAsked) {
        // If we're just starting off, we haven't asked the user for any information yet.
        // Ask the user for their name and update the conversation flag.
        case question.none:
            await turnContext.sendActivity("Let's get started. What is your name?");
            flow.lastQuestionAsked = question.name;
            break;

        // If we last asked for their name, record their response, confirm that we got it.
        // Ask them for their age and update the conversation flag.
        case question.name:
            result = this.validateName(input);
            if (result.success) {
                profile.name = result.name;
                await turnContext.sendActivity(`I have your name as ${ profile.name }.`);
                await turnContext.sendActivity('How old are you?');
                flow.lastQuestionAsked = question.age;
                break;
            } else {
                // If we couldn't interpret their input, ask them for it again.
                // Don't update the conversation flag, so that we repeat this step.
                await turnContext.sendActivity(result.message || "I'm sorry, I didn't understand that.");
                break;
            }

        // If we last asked for their age, record their response, confirm that we got it.
        // Ask them for their date preference and update the conversation flag.
        case question.age:
            result = this.validateAge(input);
            if (result.success) {
                profile.age = result.age;
                await turnContext.sendActivity(`I have your age as ${ profile.age }.`);
                await turnContext.sendActivity('When is your flight?');
                flow.lastQuestionAsked = question.date;
                break;
            } else {
                // If we couldn't interpret their input, ask them for it again.
                // Don't update the conversation flag, so that we repeat this step.
                await turnContext.sendActivity(result.message || "I'm sorry, I didn't understand that.");
                break;
            }

        // If we last asked for a date, record their response, confirm that we got it,
        // let them know the process is complete, and update the conversation flag.
        case question.date:
            result = this.validateDate(input);
            if (result.success) {
                profile.date = result.date;
                await turnContext.sendActivity(`Your cab ride to the airport is scheduled for ${ profile.date }.`);
                await turnContext.sendActivity(`Thanks for completing the booking ${ profile.name }.`);
                await turnContext.sendActivity('Type anything to run the bot again.');
                flow.lastQuestionAsked = question.none;
                profile = {};
                break;
            } else {
                // If we couldn't interpret their input, ask them for it again.
                // Don't update the conversation flag, so that we repeat this step.
                await turnContext.sendActivity(result.message || "I'm sorry, I didn't understand that.");
                break;
            }
    }
}

And here methods for validation:

// Validates name input. Returns whether validation succeeded and either the parsed and normalized
// value or a message the bot can use to ask the user again.
static validateName(input) {
    const name = input && input.trim();
    return name !== undefined
        ? { success: true, name: name }
        : { success: false, message: 'Please enter a name that contains at least one character.' };
};

// Validates age input. Returns whether validation succeeded and either the parsed and normalized
// value or a message the bot can use to ask the user again.
static validateAge(input) {
    // Try to recognize the input as a number. This works for responses such as "twelve" as well as "12".
    try {
        // Attempt to convert the Recognizer result to an integer. This works for "a dozen", "twelve", "12", and so on.
        // The recognizer returns a list of potential recognition results, if any.
        const results = Recognizers.recognizeNumber(input, Recognizers.Culture.English);
        let output;
        results.forEach(result => {
            // result.resolution is a dictionary, where the "value" entry contains the processed string.
            const value = result.resolution['value'];
            if (value) {
                const age = parseInt(value);
                if (!isNaN(age) && age >= 18 && age <= 120) {
                    output = { success: true, age: age };
                    return;
                }
            }
        });
        return output || { success: false, message: 'Please enter an age between 18 and 120.' };
    } catch (error) {
        return {
            success: false,
            message: "I'm sorry, I could not interpret that as an age. Please enter an age between 18 and 120."
        };
    }
}

// Validates date input. Returns whether validation succeeded and either the parsed and normalized
// value or a message the bot can use to ask the user again.
static validateDate(input) {
    // Try to recognize the input as a date-time. This works for responses such as "11/14/2018", "today at 9pm", "tomorrow", "Sunday at 5pm", and so on.
    // The recognizer returns a list of potential recognition results, if any.
    try {
        const results = Recognizers.recognizeDateTime(input, Recognizers.Culture.English);
        const now = new Date();
        const earliest = now.getTime() + (60 * 60 * 1000);
        let output;
        results.forEach(result => {
            // result.resolution is a dictionary, where the "values" entry contains the processed input.
            result.resolution['values'].forEach(resolution => {
                // The processed input contains a "value" entry if it is a date-time value, or "start" and
                // "end" entries if it is a date-time range.
                const datevalue = resolution['value'] || resolution['start'];
                // If only time is given, assume it's for today.
                const datetime = resolution['type'] === 'time'
                    ? new Date(`${ now.toLocaleDateString() } ${ datevalue }`)
                    : new Date(datevalue);
                if (datetime && earliest < datetime.getTime()) {
                    output = { success: true, date: datetime.toLocaleDateString() };
                    return;
                }
            });
        });
        return output || { success: false, message: "I'm sorry, please enter a date at least an hour out." };
    } catch (error) {
        return {
            success: false,
            message: "I'm sorry, I could not interpret that as an appropriate date. Please enter a date at least an hour out."
        };
    }
}

The result will look like this:

About Author

Mathias Bothe Contact me

I am Mathias, born 38 years ago in Heidelberg, Germany. Today I am living in Munich and Stockholm. I am a passionate IT freelancer with more than 14 years experience in programming, especially in developing web based applications for companies that range from small startups to the big players out there. I am founder of bosy.com, creator of the security service platform BosyProtect© and initiator of several other software projects.

No comments yet.

Leave a comment