Overview
FastGPT’s dataset search system provides four configurable retrieval modes to align query matching with specific use cases: semantic search, full-text search, hybrid search, and result reranking. Each mode offers distinct tradeoffs between semantic understanding, keyword precision, and computational efficiency, with built-in merging logic to combine results for optimized relevance.
Search Mode Reference Table
| Search Mode | Core Behavior | Key Advantages | Limitations & Notes |
|---|---|---|---|
| Semantic Search | Calculates vector distance between query and dataset content to measure similarity | Understands similar semantics, cross-language matching, multimodal support | Depends on model training quality, inconsistent accuracy, affected by keyword use and sentence completeness |
| Full-Text Search | Uses traditional full-text indexing and term matching | Optimal for locating specific terms, subjects, and predicates | No documented limitations; focused on exact keyword matching |
| Hybrid Search | Combines vector and full-text search results using the RRF formula | Delivers richer, more accurate combined results | Cannot directly filter by similarity; requires pairing with reranking for final filtering |
| Result Reranking | Uses a Rerank model to re-sort initial search results | Significantly improves retrieval accuracy in most cases; scores more reliable than vector similarity | Generates 0-1 relevance scores; works best with complete, structured queries; query rewriting is applied pre-processing |
Standard Implementation Workflow
- Submit a user query to the FastGPT dataset search endpoint.
- Select a search mode based on your retrieval goals: use semantic search for semantic matching, full-text search for exact keyword targeting, or hybrid search for balanced performance.
- If enabling reranking: a. Apply query rewriting to ensure the query has complete subjects and predicates to maximize rerank model effectiveness. b. Run the rerank model to generate 0-1 relevance scores for all initial search results.
- Merge initial vector search, full-text search, and reranked results using the RRF formula to produce the final sorted result set.
- Filter results using rerank scores if granular relevance filtering is required.
Source: FastGPT official source
Applicability and version scope
Use this page for the documented Tutorials scenario. Confirm the FastGPT, dependency, API, and deployment versions in the official source before applying a change.
Safety guardrails
Use [REDACTED_CREDENTIAL] for credentials and private data. Confirm the documented environment and version before review.
Rollback guidance
Restore the prior technical-content authority snapshot. Restore saved configuration and data snapshots, then repeat the smallest verification scenario.