What the data for this category looks like
Document data sources for new energy development include project approval documents, feasibility study reports, environmental impact assessment documents, grid connection dispatch agreements, land use pre-examination opinions, and similar materials. Multiple departments including the development and reform commission, ecological environment department, and power grid operator produce these materials in stages.
Update timing shifts irregularly alongside project progress. Updates synchronize when feasibility study reports or approval documents are revised.
Document structures vary widely. Some are plain text approval opinions. Others include feasibility study reports with tables containing installed capacity, grid connection time, and other details. Fields include project number, installed capacity (unit: MW), approval document number, original text of compliance clauses, and more. The length of a single document ranges from several pages to dozens of pages.
What constraints these characteristics impose on knowledge base retrieval and recall
Data sources are scattered across multiple functional departments. Document formats and field naming conventions differ between departments. This requires adapting to multiple structures during knowledge base import, which increases preprocessing costs.
Update rhythms are not fixed. Batch document updates occur during project adjustment phases. An incremental synchronization mechanism must be configured to ensure knowledge base timeliness.
Documents contain large numbers of professional terms and fixed-format fields. Retrieval must match both semantic meaning and structured fields to avoid generalized recall.
Some documents have long lengths. Reasonable boundaries must be set during chunking to avoid splitting key compliance clauses.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.85 | New energy development documents mostly consist of compliance content. A high matching threshold is needed to filter irrelevant results. This range balances precision and recall coverage |
limit | Top 8–12 results | Single new energy development documents have long content. Too many recall results will exceed the context window limit of FastGPT v4.17.0 |
searchMode | vector+keyword | Matching both text semantics and structured fields such as approval document numbers and installed capacity is required. This avoids biases from relying solely on vector recall |
usingReRank | Enabled | Long documents have a large number of chunks. Reranking further filters low-relevance chunks and improves the quality of final retrieval results |
datasetSearchUsingExtensionQuery | Disabled | New energy development documents have fixed terms and clear scenarios. Extended queries may introduce irrelevant content and reduce retrieval precision |
collectionIds | Bind exclusive knowledge base collections for corresponding projects | Documents from different projects do not interfere with each other. This prevents cross-project recall of documents unrelated to the current task |
Three common errors
- Symptom: Retrieval results have very low matching with the query, returning a large number of irrelevant wind power project documents. Cause: The
searchModeparameter is not configured for hybrid retrieval mode. Relying solely on vector recall fails to match structured fields, leading to deviations in semantic matching. - Symptom: Retrieval returns a timeout error, with a 504 status code displayed on the interface. Cause: The
limitparameter value is not restricted. Recalling too many long document chunks at once exceeds the system processing limit. - Symptom: Units in recall results do not match. For example, a project marked with kW is matched to a query requiring MW units. Cause: Keyword retrieval for field units is not enabled. Relying solely on vector recall fails to recognize unit differences, leading to results that do not meet business requirements.
How to confirm the configuration is correct
- Run a compliance clause query for a single new energy development document. Check whether the retrieval results include the approval document number and specific clause content of the corresponding document.
- When adjusting the
similarityparameter, verify whether the matching degree of recall results meets the business requirements for compliance. Adjust the value range to adapt to specific scenarios. - Verify that after enabling
usingReRank, the top-ranked result is the most relevant document chunk, confirming that the reranking function is active. - Check whether
collectionIdsis bound to the exclusive knowledge base collection for the current project, to avoid cross-project recall of irrelevant data.
The values provided here are common starting points. They should be measured against the reader's own samples.
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.