7.9 Knowledge Base Integration

The Website Assistant supports deep integration with the Pop Knowledge Base (KB), enabling visitors' questions to be answered using structured knowledge first. This significantly improves accuracy, consistency, and reliability. This chapter explains how to select a KB, retrieval strategies, snippet extraction, and key capabilities.

🧩 1. Enable Knowledge Base Integration

In Site Management → Website Assistant Settings, you can configure a dedicated knowledge base for each site.

Available Options:

Item	Description
Enable KB Integration	Toggle to allow assistant to use KB answers first
Knowledge Base	Select any existing Pop Knowledge Base
Retrieval Mode	BM25 / KNN Vector Search / Hybrid Search
Priority Strategy	Document Answer Priority / AI Answer Priority
Snippet Extraction	Whether to include referenced document snippets in responses

🔍 2. Document Priority Strategy

When a visitor submits a question, the system processes it using the following order:

① Text Matching (BM25)

Works best for short queries
Ideal for FAQ, structured lists, product Q&A

② Semantic Search (KNN Vector)

Understands semantic similarity
More robust for complex or flexible questions

③ Hybrid Mode (Recommended Default)

Combines BM25 with vector re-ranking
Provides the highest overall accuracy
Pop uses Hybrid by default

📘 3. Smart Snippet Extraction

When the KB finds relevant content, Pop automatically selects the most relevant sentences and returns them as a snippet.

Example:

According to the document "User Registration Guide," the registration steps include: entering email → verification → password setup.

Benefits:

More reliable (backed by your documents)
Increases user trust by showing source evidence
Can automatically include document reference info

🛠 4. KB Processing Logic (System Flow)

When the assistant receives a question, the backend follows this workflow:

User Question →
Knowledge Base Retrieval (Hybrid) →
Found Relevant Snippet?
  ├─ Yes → Return KB Answer (with snippet)
  └─ No  → Fallback (fixed response / AI response)

Explanation:

If a document is matched, the system uses the KB answer
If no match is found, fallback logic is used (explained in next chapter)

📚 5. Multiple Knowledge Base Support

Pop allows each site to select one or multiple KBs (if multi‑KB is enabled).

Typical use cases:

Product Manual KB
Privacy Policy KB
Technical Documentation KB
FAQ KB

The system merges results across KBs and automatically re-ranks them.

⚙️ 6. Advanced KB Settings

You can further fine-tune KB behavior in the backend:

① Snippet Length

Choose how many sentences to include
Helps match your document’s style

② Similarity Threshold

If similarity is below threshold, snippet is not returned
Prevents low-quality or irrelevant answers

③ Filtering Rules

Filter overly short or meaningless fragments
Exclude specific document types if needed

🧪 7. Test KB Integration

Pop provides a testing textbox in the backend so you can quickly verify how your KB responds:

Enter a user question
System shows:
- Retrieved snippet
- Similarity scores
- Retrieval method used
- Final generated answer

Useful for adjusting KB content before going live.

📝 8. Best Practices

To make your KB respond more intelligently, consider the following recommendations:

✔ Break long documents into smaller chunks

Improves matching and semantic clarity.

✔ Structure FAQ entries

Like:

Question (Q)
Answer (A)
Tags / Keywords

✔ PDFs should include headings

This helps correct segmentation and understanding.

✔ Provide examples in technical documents

Examples greatly improve retrieval quality.

📌 Summary

After enabling Knowledge Base integration, your Website Assistant will:

✔ Provide more accurate answers
✔ Offer evidence-based responses (snippet references)
✔ Align with your internal knowledge system
✔ Avoid hallucination by grounding responses in your documents

This feature is essential to making your assistant “work like a real employee.”