What the data for this use case looks like
Game event configuration data primarily comes from operation backends, player segments, monetization data, and event logs. Daily documents include event plans, reward configuration tables, launch checklists, and event post-mortem documents. Fields cover event ID, reward type, effective time window, delivery threshold, compensation rules, and more. Data update frequency is weekly or aligned with holiday event cycles. Single updates mostly adjust event rules, fine-tune reward values, or implement temporary compensation plans based on player feedback.
What constraints do these characteristics impose on re-ranking and hybrid retrieval
Because event configuration data has strong timeliness, currently active event documents must be prioritized for recall to avoid returning expired event configuration content. Documents include multiple types of text and numeric fields, so hybrid retrieval must balance keyword matching and numeric feature matching. Operations on duty and internal ticket Q&A require fast identification of issues such as abnormal compensation or configuration errors. Therefore, re-ranking and hybrid retrieval must filter content from non-current events, and adapt to the high-frequency update rhythm of weekly or holiday events to avoid recalling excessive expired data.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Filter non-current event configuration content, match relevance between event rules and ticket keywords |
limit | Top 10 entries | Control the number of results returned per retrieval, adapt to fast troubleshooting needs during on-duty operations |
searchMode | Hybrid retrieval | Combine vector retrieval and keyword retrieval, balance text keywords of event plans and numeric features of reward configurations |
usingReRank | Enabled, bound to event ID and reward value | Prioritize sorting by matching degree between effective time of current events, reward values and ticket keywords |
Three Common Misconfigurations
- Phenomenon: Empty or no matching results after enabling re-ranking. Cause: Core fields such as event ID and reward value are not included in the re-ranking weight sorting basis.
- Phenomenon: Hybrid retrieval returns a large number of expired event configuration documents. Cause: No
similaritythreshold is set, resulting in recall of offline or ended event data. - Phenomenon: Recall results in hybrid + re-ranking mode do not meet expectations. Cause: No adjustment of parameters based on the update frequency of event configurations and the timeliness of current events.
How to Confirm Proper Configuration
- Enter the event ID of the current event as the retrieval keyword, check whether the retrieval results of FastGPT v4.17.0 prioritize displaying documents for the current event.
- After enabling the re-ranking function, verify whether the returned results are sorted first by matching degree between event effective time, reward value and retrieval keywords.
- Adjust the
similaritythreshold, observe whether the number of returned results meets the fast troubleshooting needs during on-duty operations. - View retrieval logs to confirm that the re-ranking model correctly associates core fields of event configurations and does not recall content from non-current events.
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.