Index Method and Retrieval Settings

Choose the indexing method and configure retrieval strategy to optimize search results.

Table of Contents

· [Select the Index Method](#select-the-index-method)

· [Configure the Retrieval Settings](#configure-the-retrieval-settings)

· [Q&A Mode](#qa-mode)

Select the Index Method

ClickAI provides 2 indexing methods:

High Quality

· Calls an Embedding Model to create vectors for each chunk

· Enables semantic search (meaning-based retrieval)

· Pros: Higher accuracy, better contextual understanding

· Cons: Consumes embedding tokens, longer processing time

Economical

· Uses keyword indexing (inverted index) — a standard search engine technique

· No embedding model needed, no token cost

· Pros: Fast, no embedding cost

· Cons: Keyword matching only, no contextual understanding

Criteria	High Quality	Economical
Technology	Embedding vectors + keyword	Keyword indexing (inverted index)
Accuracy	High	Medium
Token cost	Yes (embedding)	No
Processing time	Longer	Faster
Semantic search	✅	❌
Multilingual search	✅	❌

💡 TIP: If you need highly accurate chatbot responses or handle multilingual content, choose **High Quality**. If speed and cost savings are priorities, choose **Economical**.

Configure the Retrieval Settings

The Knowledge Base supports 2 core retrieval techniques:

1. Semantic Retrieval: Based on vector similarity — text chunks and queries are converted into vectors and matched via similarity scoring.

2. Keyword Matching: Using an inverted index (a standard search engine technique).

Weight Settings

This feature enables you to set custom weights for semantic priority and keyword priority:

Semantic Value = 1 (Semantic Search Only)

· Activates only semantic search mode

· Uses embedding models to search deeper — even if exact query terms don't appear in the knowledge base

· Calculates vector distances to return relevant content

· Captures meaning across different languages for accurate cross-language results

Keyword Value = 1 (Keyword Search Only)

· Activates only keyword search mode

· Performs a full-text match against the knowledge base

· Suitable when users know exact terminology

· Consumes fewer resources, ideal for quick searches within large document repositories

Custom Keyword and Semantic Weights

Beyond enabling only semantic or keyword search, ClickAI provides flexible custom weight settings:

· Continuously adjust weights of both methods

· Identify the optimal weight ratio for your business scenario

Rerank Model

· Disabled by default. When enabled, a third-party Rerank model sorts the text chunks returned to optimize results

· Helps the LLM access more precise information, improving output quality

· Before enabling, go to Settings > Model Providers to configure the Rerank model API key

⚠️ IMPORTANT: If the selected embedding model is multimodal, select a multimodal rerank model (marked with a Vision icon) as well. Otherwise, retrieved images will be excluded from reranking.

📝 NOTE: Enabling Rerank will consume tokens from the Rerank model. Refer to the associated model's pricing page for details.

TopK and Score Threshold

Setting	Description	Default	Recommended
Top K	Maximum number of most similar chunks to retrieve	3	3-5
Score Threshold	Minimum similarity score for a chunk to be retrieved	0.5	0.5-0.7

· Higher Top K → retrieves more chunks, but may include less relevant content

· Higher Score Threshold → requires higher similarity, fewer but more accurate chunks returned

Q&A Mode

Q&A Mode is a special option that lets ClickAI automatically analyze documents and generate question-answer pairs. Suitable for:

· FAQ documents

· Customer support documentation

· Documents with clear Q&A structure

📖 Previous: [Chunk Settings](./03-chunk-settings.md) · Next: [Knowledge Pipeline]