Improving Your Bot's Intelligence - Handling Unknown Intents With Azure OpenAI 😎😍

Overview of the Concept

When building a conversational AI, it's crucial to have a solid plan for handling unknown intents. While natural language processing (NLP) modules can help you identify the user's intent, there may be cases when the NLP returns an "Unknown" intent, leaving your bot unable to handle the query. In such cases, your bot can respond with a message that prompts the user to ask a different question that the bot can handle.

However, taking your bot's ability to handle unknown intents to the next level requires more sophisticated responses. Instead of relying on pre-defined messages, you can send user queries to the Azure OpenAI model to generate highly intelligent and contextually relevant responses.

But there's even more you can do to improve your bot's response strategy. By routing the Azure OpenAI response to your NLP module, you can attempt to identify the user's intent again. If the NLP module finds a matching intent, your bot can respond based on that. If no matching intent is found, your bot can still respond to the user using the Azure OpenAI-generated response. This approach ensures that your bot responds to user queries in a highly intelligent and contextually relevant manner, improving its overall intelligence.

Implementing the Concept in the Bot Framework: Leveraging Conversational Language Understanding

To apply the approach discussed above in the Bot Framework, you must have already implemented the Conversational Language Understanding (CLU) concept in your bot. With CLU, you can interpret the user's intent and respond accordingly. Now, let's focus on implementing the above concept in your bot.

And make sure you have created Azure OpenAI model, to create Azure OpenAI model you can refer to this article on how to implement this concept.

To get started, you must install the preview version of "Azure.AI.OpenAI" NuGet package

Next, locate the key and endpoint in the Azure OpenAI resource you created in the Azure portal.

The below code is a C# code snippet that handles unknown user intents in a bot using Azure OpenAI. Here's what the code does,

• The method "HandleUnKnownIntent" takes a user message as input and returns a string.
• An instance of the "OpenAIClient" is created, which takes in the Azure OpenAI API endpoint and key as parameters.
• A "CompletionsOptions" object is created, which sets the prompt to the user's message and the maximum number of tokens to 1024.
• The "GetCompletionsAsync" method is called on the OpenAIClient, which takes the model ID and completions options as parameters. This method returns a response that contains the completion result.
• The method returns the first choice of the completion result, which is a text string representing the response generated by Azure OpenAI.

private async Task < string > HandleUnKnownIntent(string userMessage) {
    var client = new OpenAIClient(new Uri(endpoint), new AzureKeyCredential(key));
    var completionsOptions = new CompletionsOptions() {
        Prompt = {
                userMessage,
            },
            MaxTokens = 1024,
    };
    var completionsResponse = await client.GetCompletionsAsync(modelId, completionsOptions);
    return completionsResponse.Value.Choices[0].Text;
  }
}

Output

This bot is designed to handle food orders. Still, in a scenario where a user requests information about the weather in a city, the bot can utilize Azure OpenAI to generate a relevant and contextually appropriate response.

In conclusion, handling unknown intents is crucial for improving the intelligence of conversational AI bots. While traditional methods involve sending pre-defined responses to the user, the Azure OpenAI model provides highly sophisticated and contextually relevant responses. By routing the Azure OpenAI response to your NLP module, you can further improve your bot's ability to handle unknown intents. This approach ensures that your bot responds to user queries intelligently and reasonably, improving its overall performance. By implementing these strategies, you can take your bot's intelligence to the next level and provide a better user experience.