What the Data for This Category Looks Like
Central enterprise strategic investment departments, project companies, state-owned assets supervision authorities, and external consulting and evaluation institutions provide the core data for this category. Daily documents include annual investment plans, project proposals, feasibility study reports, due diligence reports, asset evaluation reports, post-investment evaluation reports, and other related materials. Data sources include state-owned assets supervision commission investment supervision systems, enterprise ERP and project management systems, industry market data, audit reports, and financial statements. Data is updated quarterly. Documents contain both structured project parameters and unstructured analysis content. Fields cover project cycle, investment amount, compliance requirements, and more. There is no fixed unified format, with variations across project types.
What Constraints Do These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
This category’s data characteristics create three core constraints for the retrieval and recall link. First, multi-source heterogeneous data sources require retrieval to cover both structured fields and unstructured text. Matching logic must balance system-exported data and report-style documents. Second, the quarterly update frequency requires regular incremental updates. This prevents retrieval results from lagging behind the latest investment plans and post-evaluation data. Third, project-specific document scopes require precise limitation of the retrieval set. This stops cross-project data from interfering with investment decision material review and post-evaluation analysis. Additionally, large volumes of documents generated in core time-consuming links require control of retrieval result scale. This avoids overly long contexts that disrupt subsequent processing.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Match professional documents related to investment projects, avoid irrelevant content interfering with investment decision material review |
limit | Top 8 entries | Adapt to the batch review needs of investment committee materials, keep context length within a reasonable range |
searchMode | Hybrid retrieval | Balance structured field matching and unstructured text semantic recall, cover multiple types of documents such as feasibility study reports and due diligence reports |
usingReRank | Enabled | Re-rank initial recall results, improve the display priority of high-relevance post-investment evaluation reports and asset evaluation reports |
datasetSearchUsingExtensionQuery | Enabled | Automatically expand query terms to cover retrieval scenarios for multi-source data such as annual investment plans and industry market data |
collectionIds | Bind by project category | Precisely limit the retrieval scope, avoid cross-project data interfering with current project approval and post-evaluation work |
Three Common Misconfigurations
- Issue: Retrieval results include documents from non-current approval projects, and the interface returns result sets without filtering prompts. Cause: The
collectionIdsparameter is not configured correctly, and the knowledge base collection exclusive to the current project is not bound, resulting in cross-project data being recalled. - Issue: Retrieval requests time out, and the system returns a 504 status code. Cause: The
limitparameter is not adjusted according to the quarterly updated document scale, and the number of recalled entries is set too high, causing the context length to exceed the platform's processing limit. - Issue: Uploaded video attachments from due diligence reports cannot be retrieved, and no corresponding content appears in retrieval results. Cause: The video content parsing module is not configured, only chunking and recall of text documents are supported, and non-text format attachment data cannot be processed.
How to Verify Proper Configuration
- Manually input a core calculation data fragment from a project's feasibility study report as a query term, verify that the returned results only include related documents for this project, with no investment materials from other projects.
- Check the sorting order of retrieval results, confirm that core documents such as post-investment evaluation reports and asset evaluation reports rank in the top three results.
- Upload an industry market data document updated in the current quarter, trigger an incremental update, then use relevant keywords for retrieval, confirm that the newly uploaded document is normally recalled.
- After enabling
datasetSearchUsingExtensionQuery, input simplified query terms such as "project approval", confirm that annual investment plan documents that would normally require full vocabulary matching can be recalled.
The values provided are common starting points and 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.