Amazon announced on Wednesday that it responded to consumer demand for “Alexa, but let it talk more.” digital Assistant model Be able to infer the user’s “potential goal”, and Use these questions to ask follow-up questions to users.
If you ask the new Alexa How long does it take to make an egg that is too easy, Alexa will tell you This probably requires 1 1/2 minutes-Then ask if you want to set a timer.
according to Amazon, Update Involved Algorithm tuning and deep learning model This will enable the device to According to its and user.If you ask Alexa how to cook eggs every morning and always choose to set a timer, Alexa’s Discovery model Will use Actively learn to improve its predictions and draw conclusions more accurately, do you want to know when 1 1/2 minutes.
“Amazon’s goal for Alexa is that customers should interact with her as naturally as they would with another person,” Amazon said. Blog post. “and [apps] May encounter different results, our early indicators indicate potential targets [inference] Increased customer engagement through some developer applications. “
Amazon seems to admit that the potential for potential target discovery is very, very annoying.In the early prototype User requirements For example, “Chicken Recipe”, It is reported that Alexa will Then asked: “Do you want me to play the sound of a chicken?”
In order to mitigate potential bad recommendations, Amazon implemented a The trigger model based on deep learning will “Several aspects of the conversation context, such as the text of the customer’s current conversation with Alexa and whether the customer has accepted Alexa’s multi-skilled advice in the past. “
Inferences about potential targets follow Amazon’s launch Alexa conversation–A series of deep neural networks The purpose is to manufacture It is easier for developers to integrate naturally Customize the conversation experience in the app-currently English version available in the United States.