What this category of data includes
This category of data draws primarily from contracts, work breakdown structures (WBS), customer schedules, and project management systems. Daily documents cover project plans, milestone lists, weekly reports, and risk issue lists. Data update cycles run weekly or align with individual milestone nodes. Individual documents contain structured fields such as task number, responsible person, planned start and end time, risk level, and budget proportion. Some documents are plain-text progress tracking descriptions.
What constraints these characteristics impose on knowledge base retrieval and recall
This category of data has mixed characteristics of structured fields and text descriptions, with relatively fixed update cycles. This requires knowledge base retrieval to support both structured field matching and full-text semantic recall. Multiple data sources mean recall results must integrate project documents from different systems to avoid information silos. Since project data links directly to progress, acceptance, and costs, retrieval result accuracy directly impacts decision-making, so the proportion of irrelevant recalled content must be strictly controlled. Additionally, frequently used daily content related to task decomposition and scheduling coordination accounts for a high share, so document fragments on these topics must be recalled first.
Configuration settings
This document is written based on FastGPT v4.17.0.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Adapts to the professional terminology characteristics of project management documents, balancing the completeness and accuracy of semantic recall |
limit | Top 6 entries | Adapts to the average fragment length of individual project documents, avoiding context window overflow |
searchMode | hybrid | Combines semantic recall and keyword matching to cover both structured fields and text-based project documents |
usingReRank | Enabled | Re-ranks initial recall results to prioritize content more relevant to pre-sales Q&A and technical support queries |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands user queries by adding synonyms and terms related to project management, improving recall coverage |
Three common configuration mistakes
- Failure to reproduce uploaded XLSX project document content during conversations: A prompt indicating the file cannot be read appears after initiating a query. The root cause is that structured parsing configuration for spreadsheet files is not enabled. Only plain text documents are parsed by default, and specific row and column content of the table is not extracted.
- Inability to have the large language model answer using only specified knowledge base files: Recall results include content from irrelevant knowledge bases. The root cause is that the target knowledge base collection is not bound via the
collectionIdsconfiguration, and all knowledge base resources are called by default. - Exported dataset.csv does not match the custom knowledge base format: The exported file uses a general spreadsheet template and does not match the exclusive fields of project implementation plan management. The root cause is mistakenly using the system default Q&A format template for export.
How to confirm configuration is applied correctly
- Upload a project plan XLSX file containing structured tables, initiate a query of "View the content of the 10th row", and verify whether the fields and corresponding values of that row are returned.
- Manually specify the target knowledge base collection ID, initiate a query containing keywords such as project scheduling and task decomposition, and check that recall results only include documents within this collection.
- Adjust the
similaritythreshold to 0.7 or 0.9, initiate a test query, and compare changes in the number and relevance of recall results to confirm the configuration takes effect. - Export knowledge base content, and check that the exported CSV file includes exclusive fields of project implementation plan management documents such as task number, responsible person, planned start and end time.
The values provided here 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 2026-10-19.