This section helps you with the key terms and concepts related to the Kore.ai’s XO Platform.
A virtual assistant (VA) acts as an intelligent intermediary between people, digital systems, and internet-enabled things. Using machine learning, natural language processing, and other forms of advanced software, VAs can handle complex human conversations, learn from past interactions, and improve responses over time.
Anything that a user says to the VA is an utterance. For example, if the user types Book me a flight to Orlando for next Sunday, the entire sentence is considered as the user utterance.
A few essential words in the utterance that describe what the user wants the VA to do. It is usually a combination of a verb and a noun. For example, in the above user utterance “Book me a flight to Orlando for next Sunday”, the intent is Book me a flight.
To fulfill user intent, the VA may require additional information or parameters. For example, to book a flight, it needs the source and destination city along with the travel date. In the above example, user utterance Book me a flight to Orlando for next Sunday, Orlando and next Sunday are entities.
Entities are the necessary fields, data, or words for a VA to complete the user’s request. With the required entities in hand, the Kore.ai’s XO Platform can reach out to the web service and get the specific data or perform the action as per the user intent.
For more information about parameters and user input, refer to the particular task type parameters and fields at Defining Virtual Assistant Tasks.
|Book me a flight to Orlando||Book flight||City: Orlando|
|Schedule a meeting with Bill this Sunday||Schedule a meeting||Person name: Bill
Date: April 22, 2018
|Add two bottles of red wine to the cart||Add to cart||Item: Red Wine
Natural Language Processing(NLP)
NLP enables a VA to identify the user intents; extract useful information from their utterances and map the data(entities) to the relevant tasks. It further allows VAs to process requests in text form, rather than complicated menus or programming language.
The Kore.ai’s XO Platform uses a unique NLP strategy. It combines both Fundamental Meaning and Machine Learning Engines for optimal conversational accuracy. VAs built on the XO Platform can understand and process:
- Multi-sentence messages
- Multiple intents
- Contextual references
- Patterns and idiomatic sentences, and more.
A machine’s ability to simulate human behavior and decision-making to perform tasks that normally require human-like intelligence, such as speech recognition and understanding, language translation, and more.
A machine’s ability to use algorithms, patterns, and training data to learn and find hidden insights, without being explicitly programmed.
The extraction of relevant and valuable data from a user’s utterance to complete a task. VAs can ensure they have all the data to complete tasks or prompt the user for any missing information.
Tasks refer to different types of simple and complex jobs that are designated to the VA by the developer. These tasks are performed by the VA to fulfill the user intents.
For example, for a travel virtual assistant, task names are to book tickets, find hotels, provide weather forecasts etc. that cater to different user intents.
Once the VA understands the intent, it is ready to perform a task, such as reaching out to a web service, extracting the current weather conditions report, parsing that response, and then delivering the data to the user.
Kore.ai provides five predefined task types that virtually cover all VA scenarios.
VAs can collect, modify, and post information in systems of record, like scheduling an appointment, searching for a product, or updating critical information.
VAs can deliver timely, relevant, and personalized notifications to customers and employees directly from the enterprise system by polling backend systems or by pulling information at regular intervals. Users or VA developers can configure alerts.
VAs can provide users with answers to their most commonly asked questions by querying a predefined set of information. For example, a VA can answer customer questions regarding hours of operation while also answering questions about return policies.
VAs can provide users with data from back-end systems in the form of reports. These reports are formatted and organized based on user preferences, applicable filters, and they can be downloaded for later use. For example, a VA can provide a sales manager with a report detailing the top 10 sales reps last year by region, organized from most to least sales.
VAs can handle complex multi-turn conversational experiences that replicate the natural back and forth exchanges people have every day.
Learning is about how VAs recognize new intents and entities, answer questions correctly, and identify the important aspects of a user utterance. Learning can be manual or automatic. And, like humans, VAs need to know when they are wrong and what the right action, response, or answer should be.
A form of learning in which you have input variables (X) and an output variable(Y), and you use an algorithm to learn the mapping function from the input to the output Y=f(X). Here, the VA developer acts as a teacher and has virtually full control over what the VA learns. This means that the algorithm makes predictions based on the training data provided. The VA creator or developer can manually correct these predictions by flagging the findings as correct or incorrect. Since the VA developer already knows what the VA should understand, learning can be stopped as soon as the developer decides or when the model reaches an acceptable level of performance and maturity.
A form of learning that does not require a VA developer’s supervision. Here, the VA learns from all successful utterances, meaning those utterances that were successfully recognized by the VA and the completed tasks. It uses these findings to automatically expand the model and retrain the VA, including user-provided confirmations of intents in case of conflicts. This form of training allows VAs to expand their language capabilities over a period of time and improve their accuracy while excluding failed utterances and without human intervention.
Channels refer to various communication platforms where a VA can exist such as SMS, email, mobile apps, websites, messaging apps, and more. With the Kore.ai’s XO Platform, you can design VA tasks once, and deploy across channels by merely checking a box. You can also differentiate cross-channel experiences by altering message responses or leveraging channel-specific UI elements like date selectors, carousels, and more.
Variables, Context, and Session Data
When developers create and define tasks, they can access the following:
- Session variables provided by the XO Platform.
- Custom variables that they define.
- The context that defines the scope of the variable.
For example, some API requests require you to set session variables before executing the task, or a dialog task component needs to access a session variable to transit to the next node. Dialog tasks can also access the context object with additional system variables. These session and context variables allow you to persist data and store. For example, a user’s home address for e-commerce, transportation, and home delivery-focused services is used by the VA when executing a task.