This document is written based on FastGPT v4.17.0.
What the data for this category looks like
The data for this category typically originates from content datasets, advertising revenue sharing systems, financial systems, and account backends. Daily documents cover incentive rules, settlement details, appeal records, and payment lists. Fields include creator account identifiers, settlement cycle intervals, revenue sharing calculation standards, appeal submission timestamps, payment voucher numbers, and monetary amounts. Most data updates occur monthly or are triggered by exclusive operational activities. Field formats and update schedules vary across different documents. Settlement details and payment lists have relatively high structural complexity, while incentive rules and appeal records primarily consist of unstructured text.
What constraints do these characteristics impose on knowledge base retrieval and recall?
Decentralized data sources require strictly limited retrieval scopes to prevent interference from irrelevant cross-system data. Non-real-time update frequencies require index refreshes triggered monthly or per activity cycle, to avoid retrieving expired data. Differences in document structure and fields mean retrieval modes must support both structured details and unstructured rules; a single retrieval mode cannot cover all query scenarios. The accuracy of settlement-related results directly correlates with dispute and incorrect payment risks, so low-relevance results must be filtered strictly to ensure the relevance of recalled content.
How to configure the settings
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.85–0.92 | This category of data involves settlement amounts and rules, so low-relevance results must be filtered to avoid settlement disputes |
limit | Top 8–12 entries | Balances the number of results for both settlement detail and rule queries, avoiding excessive content that increases customer support reading time |
searchMode | Semantic + full-text hybrid retrieval | Covers different query scenarios for rule-based text and structured details, adapting to diverse questions from creators and operations staff |
usingReRank | Enabled | Filters entries that are semantically similar but irrelevant, improving retrieval accuracy for settlement detail data |
datasetSearchUsingExtensionQuery | Enabled | Adapts to colloquial questions, matches standard rule text, and addresses the query needs of creators with non-standard phrasing |
collectionIds | Specify exclusive IDs for the three core collections: incentive rules, settlement details, and appeal records | Limits retrieval scope, excludes irrelevant cross-category data, and improves retrieval efficiency |
Three Common Configuration Errors
- Phenomenon: The
similarityscore of retrieval results is lower than 0.8, and full-text retrieval matching is insufficient. Cause: Thesimilaritythreshold was not adjusted based on the characteristics of the category data, and hybrid retrieval mode was not enabled, resulting in the recall of low-relevance results. - Phenomenon: The front-end interface continuously displays "Retrieving" status with no results returned. Cause: A reasonable value for the
limitparameter was not configured, or the knowledge base index was not refreshed at the monthly or activity update frequency, leading to retrieval timeout and no response. - Phenomenon: Only GPT 3.5-related options are displayed in the retrieval configuration interface, and other models cannot be selected. Cause: Multi-model adaptation was not enabled during system deployment, or the category knowledge base was not bound to an environment that supports multiple models.
How to Confirm Configuration is Complete
- Enter the FastGPT knowledge base management interface, check the
similaritythreshold configuration, and verify that the value meets the category requirements. - Initiate a settlement detail query, and check whether the number of returned results is within the range set by the
limitparameter. - Submit a colloquial question such as "Why is my settlement amount incorrect", verify whether
datasetSearchUsingExtensionQueryis effective, and confirm that standard rule content is returned. - View the retrieval log to confirm that data is only recalled from the three specified core collections, with no irrelevant cross-category results.
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.