Terms and Concepts

1. Introduction

Learn about key technical terms and concepts used in the Verax Control Center.

2. Term Summary

Term	Description
Topic	Represents the main category of the conversation, such as "Customer Support", "Technical Assistance", or "Product Inquiry”.
Subtopic	Represents a specific area within the main category, such as "Billing Issues", "Troubleshooting", "Feature Requests,".
Tag	Organizes, labels, and cross-references conversations for easier identification and retrieval in a non-hierarchical fashion.
Correctness	Represents the overall accuracy of a chatbot’s answers.
Interaction	Represents a full interaction (request and response) between the user and the chatbot.
Request	Represents a piece of text or question given to a chatbot to initiate or continue a conversation, guiding its responses and interactions.
Response	Represents the text or message that a chatbot sends back to the user, continuing the conversation by addressing the user's input or query with relevant information.
Relevance	Describes how relevant the response is to the request. The value is a score in between 0 and 100.
Groundedness	Describes how grounded a language model response is in the data. The value is a score between 0 and 100.

3. Topic and Subtopic

Topic and Subtopic are used to arrange the conversations of users and LLM models by categories and subcategories.

These terms are intuitive and provide a clear structure for understanding the hierarchy of a conversation. Using these terms ensures clarity and makes it easy to navigate the conversations based on their content or focus area.

A Topic represents the main category of the conversation. Ex: “Customer Support”, “Technical Assistance”, or “Product Inquiry”.

A Subtopic represents specific areas within the Topic (main category). Ex: “Billing Issues”, “Troubleshooting”, or “Feature Requests”. The following are examples of the Topic/Subtopic structure:

Topic: Customer Support
1. Subtopic: Billing Issues
2. Subtopic: Account Management
3. Subtopic: Service Interruptions
Topic: Technical Assistance
1. Subtopic: Troubleshooting
2. Subtopic: Software Updates
3. Subtopic: Hardware Compatibility

4. Tags

A Tag is used to organize conversations in various contexts. You can add multiple Tags to any conversation or interaction within the Vorex Control Center.

For example: Use tags such as “sales”, “customer requests”, “issues”, “returns” to add further context to interactions.

5. Correctness

Correctness represents the overall accuracy of a chatbot’s answers. The Correctness score for an interaction is represented by a number between 0 and 100.The higher the score, the more accurate the chatbot’s answers, especially in user-facing or general contexts.

6. Request

A Request is the input provided by the user to the chatbot. It can be a question, command, or any form of user message intended to initiate or continue the conversation.

For example: “What are the store hours?”, “Do you sell car batteries”, “Can I talk to a customer service agent?”

7. Response

A Response is the output generated by the chatbot in reply to the user's request. It aims to address the user's query, provide information, or guide the user to the next step in the conversation.

For example: “Our store is open from 9 a.m. to 9 p.m. Monday through Saturday”, “Yes! We sell the following car batteries:”, “I’m transferring you to a customer service agent. The estimated wait time is 5 minutes.”

8. Relevance

Relevance describes how relevant the chatbot’s response is to the user’s request. The Relevance score is calculated based on the semantic distance between the user's request and the chatbot's response. The more the chatbot respons aligns with the user's request, topic, and overall meaning, the higher the Relevance score.

9. Groundedness

Groundedness describes how closely a chatbot’s response matches the information in your knowledge base. The Groundedness score measures the semantic distance between the response and evidence that supports it in the knowledge base. The closer the answer is to the information presented in the knowledge base, the higher the Groundedness score.