What the Data for This Category Looks Like
Data for this category comes from real estate registration information systems, cadastral survey databases, housing transaction online signing systems, tax systems, and electronic certificate repositories. The update frequency is daily. Documents processed daily include registration applications, ownership certificates, survey results, tax payment vouchers, real estate ownership certificates, and real estate registers. Data fields include right holder identity information, real estate location, right type, ownership restriction status, and more. There is no uniform standardized text length; some survey result documents have relatively long lengths.
Constraints Imposed by These Characteristics on Vector Models and Indexing
Reference only; verify against the target FastGPT release and deployment documentation.
How to Set Configurations
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Ensures semantic integrity for long documents such as survey results and ownership certificates, avoiding semantic fragmentation caused by excessive chunk splitting |
| Overlap Length | 100–150 characters | Ensures semantic continuity between adjacent chunks, preventing key information across chunks from being split apart |
maxToken | 4096 token | Matches the input upper limit of mainstream open-source vector models, and carries the context information required for party building Q&A |
defaultToken | 512 token | Sets the token threshold for basic chunks, balancing single-chunk semantic integrity and indexing density |
| Similarity Threshold | 0.75–0.85 | Ensures relevance of recall results to party building learning and real estate registration questions, filtering out low-match irrelevant content |
| Single Search Limit | Top 10 entries | Limits the number of recall results, avoiding redundant information interfering with the accuracy of final Q&A |
| Vector Model Selection | Domestic open-source vector models | Meets compliance requirements for state-owned enterprise and public sector scenarios, and adapts to the semantic matching needs of ownership information for this category |
Three Common Misconfigurations
- Phenomenon: When calling a vector model accessed via Ollama, FastGPT returns a 403 access denied error, but curl tests can normally obtain vector results. Cause: The correct vector model interface access whitelist is not configured, or FastGPT’s request header parameters do not match Ollama’s verification rules.
- Phenomenon: After attempting to directly perform vector database add, delete, or modify operations via MongoDB, retrieval results become abnormal. Cause: The synchronization mapping mechanism between FastGPT and external databases is not enabled; direct operations on the vector database will destroy the association between source data and vector indexes.
- Phenomenon: Duplicate document entries still exist after knowledge base index merging. Cause: The chunk overlap length is set insufficiently, or deduplication verification based on document unique identifiers is not enabled, causing different chunks of the same source document to be indexed repeatedly.
How to Confirm Proper Configuration
- View FastGPT’s vector model call logs to confirm that the values of
defaultTokenandmaxTokenmatch the configured values. - Perform an incremental indexing test to verify whether updated source data can be synchronized to the vector index.
- Retrieve preset party building learning-related real estate registration questions, and check whether the similarity of recall results falls within the set threshold range.
- Perform an index merging operation, and check whether duplicate document entries exist in the merged index.
All provided configuration values are common starting points, and should be evaluated against local sample data.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.