What the data for this category looks like
This category of data comes from the regulatory and rule repository, higher-level law database, departmental policy document repository, and administrative regulatory document filing system. Updates occur at irregular intervals. Daily processed documents include regulatory document drafts, drafting explanations, legality review opinions, solicitation opinion summaries, and filing reports. Documents typically include fields such as document title, drafting entity, article content, review process records, and filing number. Individual document lengths vary widely, ranging from hundreds of-word review opinions to tens of thousands of-word complete drafts.
What constraints these characteristics impose on knowledge base retrieval and recall
Dispersed data sources and irregular update cycles require the retrieval link to support recall across multiple independent datasets. It must also support on-demand knowledge base updates. The wide range of document lengths, from short review opinions to long drafts, requires adaptation to text chunking and retrieval logic for different lengths. Reviews require strict comparison with higher-level laws and authority bases. Retrieval results must accurately match legal hierarchies and article associations, avoiding recall of irrelevant non-higher-level law content. Errors in the review process directly affect policy release and implementation. The recall scope and accuracy of retrieval must strictly match the core requirements of the review. No irrelevant documents should be introduced to interfere with judgment.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Administrative regulatory document legality review requires strict matching of legal articles and policy bases. This similarity range filters low-correlation content, balancing recall coverage and accuracy |
limit | Top 8–12 entries | Reviews require comparison of multi-dimensional compliance bases. This recall quantity covers core reference content while avoiding context length exceeding the model's processing limit |
usingReRank | Enabled | Reorders initial recall results by relevance, raising the display priority of highly matched documents. This adapts to the high accuracy requirements of reviews |
datasetSearchUsingExtensionQuery | Enabled | Automatically generates extended query terms based on user questions. This covers associated expressions that may be missed during reviews, improving the recall rate of higher-level laws and policy bases |
collectionIds | Bind the dataset IDs corresponding to the regulatory and rule repository, higher-level law database, and departmental policy document repository | Accurately limits the retrieval scope. Only official compliance data sources related to legality review are recalled, avoiding non-compliant documents interfering with the review |
Three Common Mistakes
- Phenomenon: Semantic retrieval fails to recall documents containing specific compliance keywords. Cause:
datasetSearchUsingExtensionQueryis not enabled. Retrieval only uses original question keywords, failing to cover synonymous or associated expressions that may be used during reviews, leading to matching failures. - Phenomenon: When retrieval results return image addresses, using a strict question-answering template prompts that no answer is found. Cause: The knowledge base configuration does not filter non-text resources. The strict question-answering template only matches text block content, unable to handle non-text retrieval results such as image addresses.
- Phenomenon: Complex compliance questions cannot receive accurate replies. Recall result quantity is insufficient or correlation is low. Cause: The
similaritythreshold is set too high, filtering some documents with acceptable but not fully matching relevance. Or thelimitvalue is too small, failing to cover sufficient reference bases.
How to Confirm the Configuration Is Properly Set Up
- Manually enter a specific question related to administrative regulatory document legality review. Check whether retrieval results include relevant content from the regulatory and rule repository and higher-level law database.
- View the knowledge base retrieval configuration panel of FastGPT v4.17.0. Confirm that
collectionIdshas been bound to the specified compliance data source datasets. - After enabling
usingReRank, compare the order of retrieval results before and after reordering. Confirm that highly matched documents are ranked higher. - Test review questions containing associated expressions. Confirm whether
datasetSearchUsingExtensionQuerytriggers extended queries and recalls relevant content.
The values provided here are common starting points and should be evaluated 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.