Enhancing Your Knowledge Database: Best Practices for Structuring and Enriching Knowledge Chunks

This article is intended to answer the most frequently asked questions on the topic, structuring and organization of content for Knowledge AI.

Knowledge AI aims to improve NLP and Conversational AI by enabling them to access and understand diverse information sources, leading to more accurate and context-aware responses. However, it is often not enough to simply upload the data, which then delivers the right results straight away. For this reason, strategies are presented below which will help to improve the retrieval of the correct results.

Consider an Orchestration Layer

Before we dive into the strategies for improving knowledge chunks, let’s talk about whether we might need an orchestration layer beforehand to ensure that only relevant user questions are allowed to pass through to the knowledge search.

What is an Orchestration Layer?

An orchestration layer is a process based design in which we prefilter the user and cater the answers based on those prefiltered criteria. For example, a question like "Are you a customer or a third party" or an API call which predefines a membership level. 

It can also include a flow which separates an FAQ question from more process driven actions, such as Bill Cancellation or other such examples. 

When Should an Orchestration Layer Be Considered?

An orchestration layer is useful if the user is only allowed to ask questions on very specific topics and other topics are to be handled on a rule-based basis. It is beneficial if the topics that are to be dealt with on a rule-based basis can be clearly distinguished from the topics that are to be dealt with via the knowledge base.

Structuring Knowledge Chunks for Clarity

A Knowledge Chunk should include clearly formulated questions along with comprehensive answers. It is essential that the answer is not only concise but also provides sufficient detail and explanations. This approach enables the AI (such as OpenAI’s ADA Text Embedding model) to accurately recognize user inquiries and deliver correct responses.

Good Example of a Well-Structured Chunk for FAQ

Question:
Is lunch included in the event ticket price?

Answer:
Lunch is not included in the event ticket price. The ticket only covers entry and access to the event activities. However, there are food options available at the venue, and you can purchase lunch from various vendors on-site. Additionally, you are welcome to bring your own meal. For more details, please check the food and beverage policies listed on our website.

Bad Example

Question:

Is lunch included in the event ticket price?

Answer:

No, lunch is not included.

Enriching Content for Enhanced AI Recognition

To enhance the AI's ability to accurately recognize inquiries, the texts within the chunks can be enriched with additional information. Focus on knowledge-based content that describes the underlying issue, avoiding legal disclaimers.

Question Enrichment

Incorporate various phrasings of a question within a single chunk to address a wider range of inquiries.

Original Question:

  • How do I reset my password?

Enriched Variants:

  • How can I change my account password?
  • What are the steps to change my password?
  • Password reset instructions?

Answer:

You can change your password under "Settings" - "User Profile" - "Change Password." Ensure that you use at least 8 characters, including at least one uppercase letter, one number, and one special character.

Bad Example

Original Question:

How do I reset my password?

Answer:

Go to the settings page and click on change password.

Balancing Technical Terms with User-Friendly Language

Utilize technical terms while providing clear explanations to prevent misunderstandings. Also, include commonly used layman’s terms to make the content more accessible to all users.

Example with Technical Terms

Question:

How can I change my bank details?

Answer:

You can update or change your bank details or your SEPA direct debit mandate through the modification service available on our website, the service app or in via the customer support

  • Website Modification Service: Visit our official website and log into your account. Navigate to the "Account Settings" or "Billing Information" section where you can securely update your bank details or SEPA mandate information.
  • Mobile App: Open our service app, log in, and go to the "Account" or "Payments" section. Here, you can change your bank account details or update your SEPA direct debit mandate with just a few taps.
  • Customer Support: If you prefer assistance, our customer support team is available to help you modify your banking information. You can contact them via phone, email, or live chat. Simply provide the necessary details, and they will process the changes for you.

Please ensure that you have all relevant information ready, such as your bank account number or SEPA mandate ID, when making these changes.

Bad Example

Question:

How to update my banking info?

Answer:

You can change your bank details on our website, via the service app or customer support.

Clarifying Abbreviations for Better Understanding

Avoid using abbreviations without clarification. If you do use them, always spell them out to enhance understanding.

Example of Avoiding Abbreviations

CT should be spelled out as Computed Tomography.

MRT should be spelled out as Magnetic Resonance Tomography, followed by a layman’s description like "tube".

You can still use the abbreviations but ensure that both the abbreviation and the spelled-out term are included.

Consolidating Duplicate Content for Efficiency

To eliminate redundancies in the knowledge database, it's crucial to consolidate similar chunks and remove duplicate content. Review multiple chunks for similar questions or answers and merge them into a single cohesive chunk.

Examples of Avoiding Duplicates

Example 1

Consider the following separate questions (each in a separate chunk) which should be combined into one chunk:

  • What are the store hours?
  • When does the store open?
  • What time does the store close?
  • What are the operating hours?

Combined Answer

Our store is open from 9:00 AM to 8:00 PM, Monday through Saturday. On Sundays, we operate from 10:00 AM to 6:00 PM.

Bad Example

Separate Questions

  • What are the store hours?
  • What time does the store close?

Individual Answers:

  • The store opens at 9:00 AM.
  • The store closes at 8:00 PM.

Conclusion

Optimizing the Knowledge Chunks in your database is essential for delivering high-quality, precise answers. By implementing structured content, enriching it with relevant details, consolidating similar inquiries, and using clear language, you’ll significantly enhance the AI’s ability to support users effectively. Start today to transform your knowledge database into a robust resource that empowers users and streamlines support processes.

 

Postscript: Managing Multiple Knowledge Sources

In scenarios where your organization has multiple locations (see chapter 5), each with varying information, it's important to manage your Knowledge Sources effectively. For example, imagine a supermarket chain with several stores, each having different opening hours and other unique traits. To handle this, you can create a distinct Knowledge Source for each store, along with one general source that contains information relevant to all locations.

By doing this, you can ensure that when users search for information, they can retrieve details specific to a particular store, while also having access to general information that applies to all stores.

Example Setup

In this setup, each Knowledge Source is assigned a unique tag. For instance, you could tag one source as "greenway" (for a specific store) and another as "general" (for company-wide information).

When using the Search-Extract-Output Node (detailed node reference here), you can apply filters to extract data only from the relevant sources. For example, you could configure the filter to pull information exclusively from sources tagged with "greenway" and "general." This way, the system retrieves accurate and store-specific information efficiently.

By utilizing tags and filtering, you maintain flexibility and precision in managing multiple Knowledge Sources, ensuring users always receive the correct information based on their query and location.


Comments

0 comments

Article is closed for comments.

Was this article helpful?
1 out of 1 found this helpful