What this category of data looks like
This category of data typically serves three stakeholder groups: game operations teams, activity planners, and data operations teams. Core daily documents fall into four categories: activity plans, reward configuration tables, launch checklists, and activity post-mortems. Data sources include operation backends, player segment data, revenue data, and activity logs. Update cycles follow weekly or holiday schedules. Documents contain fields such as reward values, activity effective time periods, player segment rules, and compensation trigger conditions. Units include in-game virtual currency, timestamps, user group tags, and other relevant types. Most documents are business tables exported directly from operation backends by operations staff, with no unified standard format.
What constraints these characteristics impose on knowledge base retrieval and recall
Business documents for this category have inconsistent formats. They also include non-text fields such as numerical values, time, and segment rules. This creates high requirements for field matching accuracy in retrieval. Frequently updated business data requires real-time indexing strategies. This prevents recalling expired activity configuration content. In operations on duty and internal ticket scenarios, most questions focus on abnormal compensation and configuration error troubleshooting. Users need to quickly locate checklists and post-mortem documents for corresponding activities. The relevance weight of recall results must be higher than that of general retrieval. Additionally, configuration parameters involved in internal tickets mostly come from tables exported from operation backends. Support for structured data retrieval matching is required.
How to configure settings
The following configurations are formulated based on the parameter rules of FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches numerical and rule fields of activity configurations, and avoids recalling irrelevant historical activity documents |
limit | Top 6 entries | Covers multiple activity documents that may be involved in a single ticket, while controlling context length |
searchMode | vector+keyword | Balances text semantic matching and exact matching of structured fields such as activity time and reward values |
usingReRank | Enabled | Re-sorts recall results by relevance, adapting to the need to quickly locate core issues in operations scenarios |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval terms based on user questions, covering professional terms in activity configurations such as "paid user acquisition activity" and "compensation distribution" |
collectionIds | Bind to a dedicated dataset for game operation activities | Avoids retrieving documents from other categories, narrowing the recall scope |
Three common misconfigurations
- The status of uploaded knowledge base files remains "Indexing" and synchronization is not completed. The cause is that the bound activity documents contain a large number of structured tables, which takes a long time to index and parse, and the batch processing parameters of the retrieval engine have not been adjusted.
- Using the [{datasetId: xxx}] format when referencing knowledge base variables does not work. The cause is that the standard variable syntax supported by the platform is not used, and an incompatible dataset ID reference format is incorrectly applied.
- Activity configuration screenshots in the knowledge base cannot be directly displayed in responses. The cause is that the retrieval association configuration for image content has not been enabled, or the screenshots have not been correctly embedded in retrievable document nodes.
How to verify configurations are set correctly
- Submit a test question such as "Is the reward configuration for last week's Spring Festival activity correct?", and check whether the recall results include the configuration table and checklist for the corresponding activity.
- View the retrieval log to confirm that the similarity scores of recall results fall within the preset
similarityrange. - Test the variable reference function to verify that structured field content in activity documents can be loaded correctly.
- Upload a test activity document, and confirm that it can be correctly retrieved after indexing is completed.
The values provided 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.