I am unable to keep the context(FAQ) of the chatbot conversation.
I have successfully integrated LUIS+QnAmaker based on this documentation.
https://learn.microsoft.com/en-us/azure/cognitive-services/qnamaker/tutorials/integrate-qnamaker-luis
I have about 3 KBs that contain the same questions but different answers. The chatbot should be able to filter to the desired FAQ and the following answers should be from the chosen FAQ from the user. Currently, it only returns the first KB for the answers unless I phrase my question this way:
Can I have the answer for FAQ1? or
Can I have the answer for FAQ2?
Hope I can have some help from the community here. Thank you!
When you are using multiple KBs with same questions, you can segregate the answers by adding different metadata to your question/answer sets. A metadata is just a collection of name / value pairs. When using the QnAMakerDialog, you have the ability to set either filtering or boosting metadata, which will be passed as part of any queries to the QnA Maker Service.
You can create your QnA Maker dialog and add a single name / value metadata pair to the MetadataFilter property on the dialog. This will then cause the dialog to only receive answers marked with the same name / value metadata.
MessagesController:
public class MessagesController : ApiController
{
internal static IDialog<object> MakeRoot()
{
var qnaDialog = new Dialogs.MyQnADialog
{
MetadataFilter = new List<Metadata>()
{
new Metadata()
{
Name = "knowledgeBase",
Value = "kb1"
}
}
};
return qnaDialog;
}
/// <summary>
/// POST: api/Messages
/// Receive a message from a user and reply to it
/// </summary>
public async Task<HttpResponseMessage> Post([FromBody]Activity activity)
{
if (activity.Type == ActivityTypes.Message)
{
await Conversation.SendAsync(activity, MakeRoot);
}
else
{
HandleSystemMessage(activity);
}
var response = Request.CreateResponse(HttpStatusCode.OK);
return response;
}
}
Similarly, if you want to use the other knowledge base, you can pass the other metadata name/value pairs.e.g, { Name="knowledgeBase",Value="kb2" }
QnAMAkerDialog:
[QnAMakerService("https://xxxx.azurewebsites.net/qnamaker/", "{EndpointKey_here}", "{KnowledgeBaseId_here}",1)]
public class MyQnADialog : QnAMakerDialog<object>
{
public override async Task NoMatchHandler(IDialogContext context, string originalQueryText)
{
await context.PostAsync($"Sorry, I couldn't find an answer for '{originalQueryText}'.");
context.Done(false);
}
public override async Task DefaultMatchHandler(IDialogContext context, string originalQueryText, QnAMakerResult result)
{
if (result.Answers.FirstOrDefault().Score > 80)
{
await context.PostAsync($"I found {result.Answers.Length} answer(s) that might help...{result.Answers.First().Answer}.");
}
else
{
await context.PostAsync($"Sorry, I couldn't find an answer for '{originalQueryText}'.");
}
context.Done(true);
}
}
Example :
Metadata tagging
When you test it, the result will be as follows:
With knowledgeBase:kb1
With knowledgeBase:kb2
Hope this helps!!!
Related
I used the "no code" way to generate a Bot in Azure and connect it to a QnA Maker knowledge base.
I have then modified the code so that the Bot uses AdaptiveCards instead of HeroCards to support Markdown formatting in MS Teams channel (format used by QnA Maker).
I am trying to add SubmitActions to these adaptive cards when there are some prompts coming from the Knowledge Base. The objective is that if the user clicks on these SubmitActions it automatically send a message back to the Bot.
Please find below the code I implemented:
// adaptive card creation
var plCardBis = new AdaptiveCard(new AdaptiveSchemaVersion(1, 0));
plCardBis.Body.Add(new AdaptiveTextBlock()
{
Text = result.Answer,
Wrap = true
});
// Add all prompt
foreach (var prompt in result.Context.Prompts)
{
plCardBis.Actions.Add(new AdaptiveCards.AdaptiveSubmitAction()
{
Title = prompt.DisplayText,
Data = prompt.DisplayText
});
}
//create the the attachment
var attachmentBis = new Attachment()
{
ContentType = AdaptiveCard.ContentType,
Content = plCardBis
};
//add the attachment
chatActivity.Attachments.Add(attachmentBis);
return chatActivity;
This works fine in the WebChat however in Teams if I click on a prompt it generates an error. Looking on internet I found that I should use an object for the Data field for Teams, instead of a simple string:
"data": {
"msteams": {
"type": "imBack",
"value": "Text to reply in chat"
},
}
Do you know how I could do that in C#? How could I update my code to add this object for the Data field? The number of Actions can vary according to the question asked by the user...
Any help would be greatly appreciated
Basically, there are two options for what you can attach to "Data" - either a plain string value, or any custom object. For your scenario, you need a custom object, so you need to define a class in your project to match what you need, something like:
public class MsTeamsDataResponseWrapper
{
[JsonProperty("msteams")]
public MsTeamsResponse MsTeamsResponse { get; set; }
}
public class MsTeamsResponse
{
[JsonProperty("type")]
public string Type { get; set; } = "imBack";
[JsonProperty("value")]
public string Value { get; set; }
}
then you'd use it like this:
...
Data = new MsTeamsDataResponseWrapper() { MsTeamsResponse = new MsTeamsResponse() { Value = prompt.DisplayText } }
...
In this case "Type" already defaults to "imBack", but you could also use it for "messageBack" at a later stage if you want to overwrite the default.
I'm creating a chatbot using LUIS template and I have an intent called StartTest with utterances like these:
Begin [TestNameEntity] on [GearEntity] and [RPMEntity]
Begin [TestNameEntity] on [GearEntity]
Begin [TestNameEntity]
If the user input matches the first utterance, I have all entities I need. However, if its input matches the second utterance, I have to ask him the RPM. And for the third utterance I have to ask the gear and the RPM.
But for the second and third utterances I need to know what the user has said for the [TestNameEntity], like this:
User: Begin test 1 on second gear
Bot: What RPM?
User: 2500
How can I achieve this?
once your Intent method is called you can use this logic
result.Entities[0].Type to get entity name
so you can iterate result.Entities to get all the entities user has entered. If any of the three entities is missing you can ask for a prompt. This is a sample code of how to achieve it.
//global variables
public string CurrentGear { get; set; }
public string CurrentRpm { get; set; }
[LuisIntent("StartTest")]
public async Task StartTestIntent(IDialogContext context, LuisResult result)
{
if (result.Entities != null && result.Entities.Count >0)
{
//sample you will get selected entity type
//var ent1=result.Entities[0].Type;
var userValues=GetUserEntities(result);
if(userValues["GearEntity"]==null)
showGearPrompt(context);
if(userValues["RPMEntity"]==null)
showRpmPrompt(context);
}
}
private string[] GetUserEntities(LuisResult result)
{
//your logic here
//return list of entities;
}
private async Task showGearPrompt(IDialogContext context)
{
PromptDialog.Text(
context: context,
resume: OnGearOptionReceivedAsync,
prompt: "please enter Gear Value",
retry: "Sorry, I didn't understand that. Please try again."
);
}
public virtual async Task OnGearOptionReceivedAsync(IDialogContext context, IAwaitable<string> gear)
{
string response = await gear;
CurrentGear = response;
}
private async Task showRpmPrompt(IDialogContext context)
{
PromptDialog.Text(
context: context,
resume: OnRpmOptionReceivedAsync,
prompt: "please enter RPM Value",
retry: "Sorry, I didn't understand that. Please try again."
);
}
public virtual async Task OnRpmOptionReceivedAsync(IDialogContext context, IAwaitable<string> rpm)
{
string response = await rpm;
CurrentRpm = response;
}
Link to how to use prompts
So...
I'm trying to use the Bot Framwork with LUIS in Swedish.
Using the samples I implemented translation of the input from Swedish to English and then called the LUIS functionality.
It worked perfect until we got some very strange intent hits from LUIS.
What we found out was that a very small spelling error (in Swedish) caused the translation to create a message that triggered wrong intent.
We can solve the problem by checking the score of the received intent, but the message back to the user "I didn't understand that" isn't especially helpful.
Running the same message through Bing Spell Check and replace the faulty text with the correct one will produce a correct behaviour (mostly).
What I would like to do is to use the result from the Spell Check to ask the user if the text he/she should be replace with the result from Bing.
Now the problem: I can't find a proper way to implement the dialog to the user. (If possible, I would like to avoid using the PromptDialog.Confirm since it is tricky to localize)
What I have now (that doesn't work) is roughly:
if (activity.Type == ActivityTypes.Message)
{
correctedText = await sc.BingSpellCheck(activity.Text, spellValues);
if (spellValues.Count > 0)
{
// Ask the client if the correction is ok
await Conversation.SendAsync(activity, () => new CorrectSpellingDialog());
}
Translate.Current.ToEnglish(activity.Text, "en");
await Conversation.SendAsync(activity, () => new MeBotLuisDialog());
}
What I would like here is to create a CorrectSpellingDialog() that just returns true or false, nad if it is true I will call the ...MeBotLuisDialog().
Sorry for all the text but it's a long problem :-)
Any takers on this?
(The other solution I had was to create an Intent "SpellCheckError" that is trigged from the Bing Spell Check and the in the intent send a message with the corrected message back to the bot (even though I don't know I that is doable in a proper way))
// Tommy
To enable Bing Spell Check API in your bot, you will first get the key from Bing service, then in the Web.config file add the BingSpellCheckApiKey and together enable IsSpellCorrectionEnabled for example:
<appSettings>
<!-- update these with your BotId, Microsoft App Id and your Microsoft App Password-->
<add key="BotId" value="YourBotId" />
<add key="MicrosoftAppId" value="" />
<add key="MicrosoftAppPassword" value="" />
<add key="BingSpellCheckApiKey" value="YourBingSpellApiKey" />
<add key="IsSpellCorrectionEnabled" value="true" />
</appSettings>
Then you can create a component for your bot to use Bing Spell Api, for more information, you can refer to Quickstart for Bing Spell Check API with C#. Here you can a service in your app for example like this Service.
Then in MessagesController, check the spell first before sending message to dialog. and since your want to show a confirm dialog to user if the text should be replace with the result from Bing, you can send a FormFlow first to let user to have a choice. For example:
Code of MessagesController:
private static readonly bool IsSpellCorrectionEnabled = bool.Parse(WebConfigurationManager.AppSettings["IsSpellCorrectionEnabled"]);
private BingSpellCheckService spellService = new BingSpellCheckService();
/// <summary>
/// POST: api/Messages
/// Receive a message from a user and reply to it
/// </summary>
public async Task<HttpResponseMessage> Post([FromBody]Activity activity)
{
if (activity.Type == ActivityTypes.Message)
{
if (IsSpellCorrectionEnabled)
{
try
{
var text = await this.spellService.GetCorrectedTextAsync(activity.Text);
if(text != activity.Text)
{
//if spelling is wrong, go to rootdialog
await Conversation.SendAsync(activity, () => new Dialogs.RootDialog(activity.Text, text));
}
else
{
//if right, direct go to luisdialog
await Conversation.SendAsync(activity, () => new Dialogs.MyLuisDialog());
}
}
catch(Exception ex)
{
Trace.TraceError(ex.ToString());
}
}
else
{
await Conversation.SendAsync(activity, () => new Dialogs.MyLuisDialog());
}
}
else
{
HandleSystemMessage(activity);
}
var response = Request.CreateResponse(HttpStatusCode.OK);
return response;
}
Code of my RootDialog:
[Serializable]
public class RootDialog : IDialog<object>
{
private string otext;
private string ntext;
public RootDialog(string oldtext, string newtext)
{
otext = oldtext;
ntext = newtext;
}
public async Task StartAsync(IDialogContext context)
{
context.Wait(MessageReceivedAsync);
}
public async Task MessageReceivedAsync(IDialogContext context, IAwaitable<IMessageActivity> result)
{
var message = await result;
var form = new FormDialog<Confirmation>(new Confirmation(otext, ntext),
Confirmation.BuildForm, FormOptions.PromptInStart, null);
context.Call(form, this.GetResultAsync);
}
private async Task GetResultAsync(IDialogContext context, IAwaitable<Confirmation> result)
{
var state = await result;
await context.Forward(new MyLuisDialog(), null, context.Activity, System.Threading.CancellationToken.None);
}
}
[Serializable]
public class Confirmation
{
public string Text { get; set; }
private static string otext;
private static string ntext;
public Confirmation(string oldtext, string newtext)
{
otext = oldtext;
ntext = newtext;
}
public static IForm<Confirmation> BuildForm()
{
return new FormBuilder<Confirmation>()
.Field(new FieldReflector<Confirmation>(nameof(Text))
.SetType(null)
.SetDefine(async(state, field) =>
{
field
.AddDescription(otext, otext)
.AddTerms(otext, otext)
.AddDescription(ntext,ntext)
.AddTerms(ntext, ntext);
return true;
}))
.Build();
}
}
We have a bot that will be collecting information and we would like to utilize FormFlow. Using the custom prompter we can customize the outgoing messages, but is there any similar facility to let us intercept incoming messages before they hit the recognizers? The specific use case is based on the user input, we may want to immediately exist out of a flow and redirect to a different dialog.
You can use business logic to process user input when using FormFlow, where you can immediately exit out of a form flow and redirect to a different dialog.
Since the validate function cannot pass context, you can store the context in a form variable that is populated when constructed.
public MyFormClass(IDialogContext context)
{
Context = context;
}
public IDialogContext Context { get; set; }
public int IntegerField { get; set; }
Later, call the validate function for a specific field. Here, you can use the stored context to start a new dialog.
return new FormBuilder<MyFormClass>()
.Field(nameof(IntegerField),
validate: async (state, value) =>
{
var result = new ValidateResult { IsValid = true };
if (state.IntegerField > 10)
{
await state.Context.Call(new Dialog(), Dialog.ResumeMethod);
return result;
}
else
{
return result;
}
}
)
.Build();
Note:
Even though the first return statement will never be reached, it is required to avoid throwing an error.
You may need to implement additional steps to properly manage the bot's dialog stack.
Even though the first return statement will never be reached, it is required
i'm trying to wrap my head around how to enterprise up my code: taking a simple routine and splitting it up into 5 or 6 methods in 3 or 4 classes.
i quickly came up three simple examples of code how i currently write it. Could someone please convert these into an MVC/MVP obfuscated version?
Example 1: The last name is mandatory. Color the text box red if nothing is entered. Color it green if stuff is entered:
private void txtLastname_TextChanged(object sender, EventArgs e)
{
//Lastname mandatory.
//Color pinkish if nothing entered. Greenish if entered.
if (txtLastname.Text.Trim() == "")
{
//Lastname is required, color pinkish
txtLastname.BackColor = ControlBad;
}
else
{
//Lastname entered, remove the coloring
txtLastname.BackColor = ControlGood;
}
}
Example 2: The first name is optional, but try to get it. We'll add a bluish tint to this "try to get" field:
private void txtFirstname_TextChanged(object sender, EventArgs e)
{
//Firstname can be blank.
//Hint them that they should *try* to get it with a bluish color.
//If they do enter stuff: it better be not all spaces.
if (txtFirstname.Text == "")
{
//Nothing there, hint it blue
txtFirstname.BackColor = ControlRequired;
}
else if (txtFirstname.Text.Trim() == "")
{
//They entered spaces - bad user!
txtFirstname.BackColor = ControlBad;
}
else
{
//Entered stuff, remove coloring
txtFirstname.BackColor = SystemColors.Window;
}
}
Example 3 The age is totally optional. If an age is entered, it better be valid:
private void txtAge_TextChanged(object sender, EventArgs e)
{
//Age is optional, but if entered it better be valid
int nAge = 0;
if (Int32.TryParse(txtAge.Text, out nAge))
{
//Valid integer entered
if (nAge < 0)
{
//Negative age? i don't think so
txtAge.BackColor = ControlBad;
}
else
{
//Valid age entered, remove coloring
txtAge.BackColor = SystemColors.Window;
}
}
else
{
//Whatever is in there: it's *not* a valid integer,
if (txtAge.Text == "")
{
//Blank is okay
txtAge.BackColor = SystemColors.Window;
}
else
{
//Not a valid age, bad user
txtAge.BackColor = ControlBad;
}
}
}
Every time i see MVC code, it looks almost like random splitting of code into different methods, classes, and files. i've not been able to determine a reason or pattern to their madness. Without any understanding of they why it's being one some way, it makes no sense. And using the words model, view, controller and presenter, like i'm supposed to know what that means, doesn't help.
The model is your data.
The view shows data on screen.
The controller is used to carry out
the users actions
And oranges taste orangy.
Here's my attempt at splitting things up in order to make the code more difficult to follow. Is this anywhere close to MVC?
private void txtFirstname_TextChanged(object sender, EventArgs e)
{
FirstnameTextChangedHandler(sender, e);
}
private void FirstnameTextChangedHandler(sender, e)
{
string firstname = GetFirstname();
Color firstnameTextBoxColor = GetFirstnameTextBoxColor(firstname);
SetFirstNameTextBoxColor(firstnameTextBoxColor);
}
private string GetFirstname()
{
return txtFirstname.Text;
}
private Color GetFirstnameTextBoxColor(string firstname)
{
//Firstname can be blank.
//Hint them that they should *try* to get it with a bluish color.
//If they do enter stuff: it better be not all spaces.
if (firstname == "")
{
//Nothing there, hint it blue
return GetControlRequiredColor();
}
else if (firstname.Trim() == "")
{
//They entered spaces - bad user!
return GetControlBadColor();
}
else
{
//Entered stuff, remove coloring
return GetControlDefaultColor();
}
}
private Color GetControlRequiredColor()
{
return ControlRequired;
}
private Color GetControlBadColor()
{
return ControlBad;
}
private Color GetControlGoodColor()
{
return ControlGood;
}
//am i doin it rite
i've obfuscated the code, but it's still altogether. The next step in the MVC obfuscation, i gather, is to hide the code in 3 or 4 different files.
It's that next step that i don't understand. What is the logical separation of which functions are moved into what other classes? Can someone translate my 3 simple examples above into full fledged MVC obfuscation?
Edit: Not ASP/ASP.NET/Online. Pretend it's on a desktop, handheld, surface, kiosk. And pretend it's language agnostic.
The purpose of MVC/MVP patterns is not obfuscation, but separation of concerns. Obfuscation is to (conceal the) intended meaning in communication, making communication confusing, intentionally ambiguous, and more difficult to interpret: ref. The use of patterns is to make the code cleaner and more understandable. I suggest you start out by reading the wikipedia entries on MVC and MVP.
Both patterns are ways of structuring your code so that your application is broken up into elements that carry out specific purposes that have clearly defined interaction boundaries. Rather than having code that specifically addresses business concerns, input/output handling, and presentation throughout the various classes of the application, these concerns are separated and isolated in the various architectural components. These architectural elements are insulated from one another by the interaction boundaries (interfaces) making them more independent of one another and easier to modify without affect the application as a whole.
The main idea I have when implementing MVC for Windows Forms is that I want to have unit tests for my model and my controller. In order to achieve that, my controller should not know anything about the views using it, and so any notifications that should be handled on UI level are implemented as events. In your example, my controller would look something like this:
class Controller
{
// This is the model we are operating on
private Model model_;
public enum Status
{
Normal,
Required,
Good,
Bad
}
public delegate void FirstNameStatusChangedDelegate(Status newStatus);
public event FirstNameStatusChangedDelegate FirstNameStatusChangedEvent;
public string FirstName
{
get { return model_.FirstName; }
set
{
if (value == "")
RaiseFirstNameStatusChanged(Status.Required);
else if ( value.Trim() == "" )
RaiseFirstNameStatusChanged(Status.Bad);
else
{
model_.FirstName = value;
RaiseFirstNameStatusChanged(Status.Normal);
}
}
}
private void RaiseFirstNameStatusChanged(Status newStatus)
{
if ( FirstNameStatusChangedEvent != null )
FirstNameStatusChangedEvent(newStatus);
}
}
And the view would provide handlers for the FirstNameStatusChanged event:
class View : Form
{
private Controller controller_;
private static readonly Dictionary<Controller.Status, Color> statusColors_ = new Dictionary<Controller.Status, Color>
{
{Controller.Status.Normal, SystemColors.Window},
{Controller.Status.Required, ControlRequired},
{Controller.Status.Good, ControlGood},
{Controller.Status.Bad, ControlRed}
};
public View(Controller controller)
{
InitializeComponent();
controller_ = controller;
contoller_.FirstNameStatusChangedEvent += OnFirstNameStatusChanged;
}
private void txtFirstname_TextChanged(object sender, EventArgs e)
{ controller_.FirstName = txtFirstName.Text; }
private void OnFirstNameStatusChanged(Controller.Status newStatus)
{ txtFirstName.BackColor = statusColors_[newStatus]; }
}
Most of what you doing in your code belongs to the Controller class since it describes the the logic. Your View should just describe UI and give easy access to UI components. Model class should describe your data model.
The idea is simple: Controller does everything, but it has to know about the View and the Model. For example as View is initialized, Controller sets up all the logic ( kinda what you already doing). As Model is assigned to the Controller - it sets the values into appropriate UI controls and does the same to retrieve data and return is as Model.
So basically you give your data model class to the controller, it does the editing and returns your data as model class again.
It would be very hard to follow MVC in classic ASP.NET if possible, so I will reply based on MVP.
On your first example, you are trying to do a validation. Validating a surname is the responsibility of Presenter. Showing the field red is the responsibility of View. So, your view class would be like this:
private void Page_Load()
{
this._presenter = new Presenter();
}
private void txtLastname_TextChanged(object sender, EventArgs e)
{
txtLastName.BackColor = presenter.IsLastnameValid(txtLastName.Text) ?
ControlGood : ControlBad;
}
And your presenter class would be something like this:
public Presenter()
{
public bool IsLastNameValid(string lastname)
{
return string.IsNullOrEmpty(lastname);
}
}
Last name is your model here.
Please note that I prepared this classes only for showing how would you form an MVP structure. In real world, there are lots of better ways to do validation. Normally you would use this approach for your business instead of validation.
Ian,
If you want the controls to validate immediately, you need to use javascript or jQuery. This is also true for classic ASP.NET. Since you are using Code Behind methods, I assume that your validation waits for a postback.
The following examples are from the NerdDinner project. NerdDinner is an open source project that serves as an example of ASP.NET MVC architecture. The authors have graciously provided a tutorial with it, available at http://nerddinnerbook.s3.amazonaws.com/Intro.htm
When a form is submitted in ASP.NET MVC, it enters the corresponding controller as a FormCollection object:
[AcceptVerbs(HttpVerbs.Post)]
public ActionResult Edit(int id, FormCollection formValues) {
Dinner dinner = dinnerRepository.GetDinner(id);
try
{
UpdateModel(dinner);
dinnerRepository.Save();
}
catch
{
ModelState.AddModelErrors(dinner.GetRuleViolations())
}
return RedirectToAction("Details", new { id = dinner.DinnerID });
}
UpdateModel takes the form values and attempts to stuff them into the dinner object. The dinner object looks like this:
public partial class Dinner {
public bool IsValid {
get { return (GetRuleViolations().Count() == 0); }
}
public IEnumerable<RuleViolation> GetRuleViolations() {
yield break;
}
public IEnumerable<RuleViolation> GetRuleViolations() {
if (String.IsNullOrEmpty(Title))
yield return new RuleViolation("Title is required", "Title");
if (String.IsNullOrEmpty(Description))
yield return new RuleViolation("Description is required", "Description");
if (String.IsNullOrEmpty(HostedBy))
yield return new RuleViolation("HostedBy is required", "HostedBy");
if (String.IsNullOrEmpty(Address))
yield return new RuleViolation("Address is required", "Address");
if (String.IsNullOrEmpty(Country))
yield return new RuleViolation("Country is required", "Address");
if (String.IsNullOrEmpty(ContactPhone))
yield return new RuleViolation("Phone# is required", "ContactPhone");
if (!PhoneValidator.IsValidNumber(ContactPhone, Country))
yield return new RuleViolation("Phone# does not match country", "ContactPhone");
yield break;
}
partial void OnValidate(ChangeAction action) {
if (!IsValid)
throw new ApplicationException("Rule violations prevent saving");
}
}
Notice the IsValid method and the RuleViolations enumerator. If everything is set up properly, all you have to do is define your validations in here, and ASP.NET MVC will take care of the rest for you.
The final validated result looks like this:
I encourage you to get the NerdDinner application and tutorial at http://nerddinner.codeplex.com/