What the Data for This Category Looks Like
This category serves open source office teams, project maintainers, and legal staff. Data comes primarily from code repositories, dependency lists, license texts, and contribution records. Maintained documents include LICENSE files, NOTICE files, dependency license lists, and copyright statements. LICENSE files contain license terms and copyright ownership notices. NOTICE files compile copyright information for third-party dependencies. Dependency license lists record dependency names, versions, license types, and copyright holders in a structured format. Contribution records label code committers and their corresponding copyright ownership. Data updates trigger with each version release or every dependency change.
What Constraints Do These Characteristics Impose on the Knowledge Base Retrieval and Recall Process
This category’s data characteristics require retrieval and recall workflows to adapt to both structured and unstructured documents. The structured nature of dependency license lists means retrieval must support filtering by dependency name and version fields to avoid matching compliance information from unrelated dependencies. Data updates occur frequently. The retrieval link must support incremental synchronization to ensure the knowledge base obtains the latest content quickly after each version or dependency change. Semantic differences across documents are significant. For example, LICENSE files use specialized terms while NOTICE files contain summarized information. The retrieval model must accurately distinguish compliance needs across different scenarios, while avoiding recalling irrelevant code commit records. Additionally, the accuracy of compliance content directly impacts business risk. Retrieval and recall results must have high confidence, and must not miss critical license terms or copyright statements.
How to Configure Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Balances precision and recall range for semantic similarity matching of license texts and dependency statements, and adapts to the specialized semantic features of compliance content |
limit | Top 10 results | Covers common license types and dependencies, while avoiding excessive results that exceed the context window limit of FastGPT v4.17.0 |
searchMode | vector_search | Prioritizes vector retrieval to meet the need for specialized semantic matching of license texts, improving recall accuracy for compliance content |
usingReRank | Enabled | Re-ranks recall results to return the most relevant license compliance content first, reducing interference from irrelevant results |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands query terms—for example, expanding "GPL license" to keywords corresponding to the license text—to improve cross-document recall coverage |
collectionIds | Fill in the actual deployed license compliance knowledge base collection ID | Restricts the retrieval scope to only compliance documents such as LICENSE and NOTICE files, avoiding mixing in other unrelated project data |
Three Common Misconfigurations
- Symptom: Only the first question in a workflow triggers knowledge base citations, with no knowledge base results for subsequent questions. Cause: Session-level dataset ID persistence is not configured, or the retrieval node does not bind the correct reference path for global variables.
- Symptom: After importing a bilingual knowledge base, the large language model does not read the corresponding Chinese content. Cause: Retrieval configuration does not adapt to vector generation for Chinese semantics, or
datasetSearchUsingExtensionQuerydoes not expand Chinese keywords. - Symptom: After associating a knowledge base, the model prompts that no specific knowledge base content is provided. Cause:
collectionIdsis empty or filled incorrectly, and the retrieval scope does not cover the target license compliance document collection.
How to Verify Proper Configuration
- Manually trigger the knowledge base retrieval node, input "MIT license terms", and check whether the returned results include the core content of the corresponding LICENSE file.
- View the retrieval node's running logs to confirm that the number of recall results matches the value set for the
limitconfiguration. - Test multi-turn conversations, input "Apache 2.0 license" and "Current project's dependency license list" in sequence, to verify that subsequent questions can correctly recall relevant compliance content.
- Check the knowledge base's synchronization status to confirm that the latest dependency license lists and NOTICE files have completed incremental import.
The values provided 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 2026-10-19.