What This Category of Data Includes
Daily documents for this category cover activity plans, reward configuration tables, launch checklists, and activity review materials. Data sources include operation backends, player segment data, payment statistics, and activity operation logs. Update frequency is weekly, or temporarily adjusted before holiday events. Document fields include activity ID, time range, reward value and unit, and trigger conditions. Reward configuration tables specify item type and distribution quantity. Activity logs record each configuration’s execution and exception details.
Constraints for Knowledge Base Retrieval and Recall
Activity configuration data has high structural complexity. It includes numeric reward parameters and time-based constraint fields. Retrieval must match fields and units exactly. This avoids incorrect configuration recalls caused by fuzzy matching. Data update frequency shifts with activity cycles. The system must support incremental sync and temporary updates. This ensures pre-event configuration documents stay current. Incorrect reward distribution triggers player complaints and revenue fluctuations. Recall result confidence must align with business risk levels. Multi-source data has format differences. The retrieval process must support table formats exported from operation backends and structured log text.
Configuration Settings
This configuration is based on FastGPT v4.17.0. Specific parameter settings are listed below:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches structured fields of activity configurations. Balances precision and recall coverage. Prevents low-relevance results from being included. |
limit | Top 10 entries | Controls the number of documents retrieved per request. Fits the short-context interaction needs of operation scenarios. |
searchMode | Exact matching | Prioritizes field keyword matching for structured configuration tables and checklists. Improves matching accuracy. |
usingReRank | Enabled | Re-sorts recall results. Raises the priority of high-value reward configuration documents. Reduces exposure of incorrect configurations. |
datasetSearchUsingExtensionQuery | Disabled | Activity configuration queries are usually clear. No extra query expansion is needed to avoid interfering with core matching logic. |
collectionIds | Activity configuration-specific collection ID | Limits retrieval scope. Excludes unrelated game development and art resource documents. Improves retrieval efficiency. |
Three Common Configuration Mistakes
- Symptom: The number of retrieved results is far lower than expected, or no matching activity configuration documents are returned. Cause:
collectionIdsis not set correctly to limit retrieval scope. This causes retrieval to cover non-activity configuration documents. - Symptom: Image link content stored in the knowledge base cannot be output in conversations. Cause: Knowledge base image parsing is not enabled, or image URLs are not properly vector-embedded. This prevents associated display.
- Symptom: Retrieval and recall latency increases significantly when entering long activity rule descriptions. Cause: Query text is not truncated reasonably, or the
limitparameter is not configured to limit recall quantity. This leads to increased redundant computation.
How to Verify Successful Configuration
- Run test queries. Enter common activity configuration questions. Check if returned results include target documents such as activity plans and reward configuration tables. Verify retrieval scope is correct.
- Check the order of retrieval results. Confirm high-priority reward configuration documents appear at the top. Verify the effect of the
usingReRankconfiguration matches expectations. - Import a test activity configuration PPT. Confirm images and text are correctly added to the knowledge base. Verify parsing configuration is active.
- Adjust the
similarityparameter range. Test recall results under different thresholds. Confirm matching precision meets business risk requirements.
The values provided are common starting points. They 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.