Entity Recognition

What is Entity Recognition?

Entity recognition identifies specific products, tickers, crypto assets, and other named items in your queries. This allows the system to understand what you're asking about and retrieve more targeted information.

Product and Ticker Detection

When you mention a Bitwise product or ticker, the system recognizes it and can filter results to documents that specifically discuss that product.

Recognized Entities

ETF Tickers: BITB, ETHW, BITQ, BWEB, etc.
Index Names: Bitwise 10, BITWISE10, DeFi Index, etc.
Crypto Assets: Bitcoin, BTC, Ethereum, ETH, Solana, etc.
Fund Names: Bitwise Bitcoin ETF, Bitwise Web3 ETF, etc.

Examples

Query: “What is BITB's expense ratio?”

Detected: BITB (Bitwise Bitcoin ETF)

Query: “Tell me about the Bitwise 10 index”

Detected: Bitwise 10 → BITWISE10

Query: “How does Ethereum staking work?”

Detected: Ethereum → ETH

How Entities Improve Results

When the system detects an entity in your query, it can:

  • Expand queries: Add product context from the product catalog
  • Filter chunks: Only return chunks mentioning the specific product
  • Fetch live data: Trigger MCP calls for current prices/data
  • Improve precision: Avoid returning info about similar but different products

Entity Overlap Filtering

The entity overlap filter ensures that when you ask about a specific product, the retrieved chunks actually mention that product. This prevents the system from confusing similar products.

Entity Overlap Filter

When enabled, chunks that don't mention the detected product are filtered out. This prevents cross-contamination between similar products.

Applied to: Factual queries with specific product mentions
Not applied to: Exploratory queries, comparisons, general questions
Fallback: Keeps top 3 chunks by relevance if filter too aggressive

When Entity Filtering Helps

Query: “What is BITB's AUM?”

Without filter: May return chunks about ETHW, BITQ, or other ETFs
With filter: Only returns chunks specifically mentioning BITB

Crypto Asset Normalization

Crypto assets can be referred to by many names. The system normalizes these to canonical symbols for consistent retrieval.

User InputNormalized To
Bitcoin, BTC, bitcoinBTC
Ethereum, ETH, EtherETH
Solana, SOLSOL
Cardano, ADAADA
Dogecoin, DOGE, dogeDOGE

Troubleshooting Entity Issues

Common Entity Issues

  • Entity not recognized: Try using the official ticker symbol (e.g., “BITB” instead of “the Bitcoin ETF”)
  • Filter too aggressive: Results filtered out because they don't mention the exact product. Try rephrasing or disabling the entity filter.
  • Wrong product returned: Similar names may confuse the system. Be specific with ticker symbols.
  • Comparison queries filtered: Entity filter is disabled for comparison queries, but if it's not detected as a comparison, try rephrasing (e.g., “Compare BITB and ETHW”).

Adjusting Entity Filtering

In the Admin Console, you can configure entity filtering behavior:

  • Enable/disable: Turn entity overlap filtering on or off globally
  • Fallback count: Minimum chunks to keep when filter is too aggressive
  • Query intent override: Filtering only applies to factual queries by default

Related Articles