What this category’s data looks like
Data for this category primarily comes from daily outputs and compliance-related documents generated during game development and publishing. This includes game text, plot scripts, art resource descriptions, and internal documents such as content self-inspection forms, sensitive content modification records, user agreements, and compliance review minutes. Data updates follow game version iterations or event launch schedules, with one update per version or event. Most documents are a mix of structured and semi-structured content, containing text paragraphs, field annotations, review comments, and other elements. There is no unified fixed format. Some documents include associated resource paths and version numbers.
Constraints imposed by these characteristics on reranking and hybrid retrieval
Documents for game content compliance self-inspection are mostly non-standardized content updated with version iterations. They must be linked to regulatory requirements for the corresponding version. Therefore, reranking and hybrid retrieval must adapt to frequently updated datasets. Field differences across documents are significant. During retrieval, content strongly related to the current compliance scenario must be prioritized. This avoids interference from irrelevant documents. There is a need for associated retrieval of art resources and text. This requires the retrieval pipeline to support mixed matching of both text and metadata. Data increases from version updates. The response speed of retrieval and reranking workflows must adapt to the fast-paced business rhythm. This avoids retrieval delays caused by growing data volume. Compliance scenarios have higher requirements for retrieval result relevance. Sensitive content must not be missed. Irrelevant information must not be introduced. This places stricter requirements on the parameter accuracy of retrieval and reranking.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
usingReRank | Enabled | Game compliance self-inspection requires precise matching of sensitive content and regulatory requirements. Reranking filters low-relevance retrieval results to improve compliance judgment efficiency |
similarity | 0.70–0.85 | Compliance scenarios require strict control of retrieval relevance thresholds to avoid missing sensitive documents or introducing irrelevant content, calibrated based on actual testing |
limit | Top 8–12 results | Balances coverage of compliance documents and retrieval response speed. The single return limit adapts to dataset scale after version updates |
searchMode | hybrid | Adapts to the characteristics of game compliance data including text, metadata, and resource association information. Hybrid retrieval supports simultaneous matching of semantics and attribute tags |
| Rerank model call timeout | 30–60 seconds | Compliance documents are mostly long text, so sufficient time must be reserved for rerank model processing to avoid interrupting the retrieval workflow due to timeout |
Three Common Mistakes
- After FastGPT v4.17.0 is deployed offline, configuring the rerank model and clicking model test results in a request error. The cause is that the offline environment did not correctly load the local dependency packages for the rerank model, or the configuration path does not point to the local model file.
- Retrieval speed is too slow after enabling reranking for knowledge base question answering. The cause is that the number of
limitentries set for a single retrieval is too high, and compliance documents are mostly long text. The time taken by the rerank model to process long text increases significantly as the number of results increases. - Retrieval results are empty or have insufficient relevance after configuring the rerank model. The cause is that the
similaritythreshold is set too high, causing eligible retrieval results to be filtered out, orsearchModedoes not adapt to the mixed data structure of compliance documents.
How to Confirm Configuration is Complete
- Access the knowledge base retrieval configuration page, confirm that the
usingReRankswitch is enabled. - Input a compliance-related test question, review the sorting logic of retrieval results, and confirm that highly relevant documents are displayed first.
- Simulate dataset growth after a version update, run batch retrieval tests, and confirm that retrieval response time meets business rhythm requirements.
- Call the model test interface, verify that the rerank model can return results normally, with no timeout or error prompts.
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 2026-10-19.