What the Data for This Category Looks Like
The data for this category primarily comes from advertising platform backends, attribution analysis systems, CRM lead databases, and landing page data collection tools. Commonly processed documents include daily and weekly ad delivery reports, single price adjustment operation records, material A/B test result tables, and delivery effect optimization review notes. Most fields cover delivery-related metrics such as impressions, clicks, conversion costs, and budget limits. Data updates follow a daily cadence for most content. Core real-time adjustment data, such as individual bid records and real-time landing page conversion data, is updated in real time.
Constraints Imposed on Knowledge Base Retrieval and Recall
Multi-source and heterogeneous data sources require the knowledge base to support unified indexing and cross-system data recall, to eliminate data silos. The mixed daily and real-time update rhythm requires the retrieval process to support incremental synchronization and fast real-time data recall, to maintain result timeliness. Document types range from structured tables to unstructured notes, so different chunking and retrieval matching logic must be supported to ensure accurate recall of content in all formats, including price adjustment records and review notes. Differences in delivery-related metric fields require that specific fields can be specified as recall criteria during retrieval, to avoid interference from irrelevant data.
Configuration Settings
This guidance is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Aligns with the field density and recall accuracy requirements of delivery data, to avoid recalling too many irrelevant historical price adjustment records or low-relevance review content |
limit | Top 10 entries | Controls the number of results returned per recall, adapts to the usage habits of ad optimizers who browse quickly, and avoids result overload |
usingReRank | Enabled | For multi-source heterogeneous delivery data, re-ranking improves the sorting priority of highly relevant content, solving the problem of mixed initial recall results |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands query terms based on user delivery questions, supplementing industry terminology such as "conversion cost" and "budget allocation" to improve recall coverage |
collectionIds | Specified by business domain | Limits the retrieval scope to the knowledge base collection dedicated to ad delivery effect optimization, to avoid recalling document data from other categories |
searchMode | Hybrid retrieval | Adapts to structured price adjustment records and unstructured review documents, balancing exact matching and semantic matching |
Three Common Configuration Errors
- Enabling
usingReRankmay result in empty retrieval returns. The symptom is that no valid content snippets are displayed after running a retrieval. The cause is that the re-ranking relevance threshold is set too high, filtering out all valid content from the initial recall results. - A locally deployed knowledge base cannot create more than the allowed number of collections. The symptom is that a creation failure prompt appears in the interface. The cause is that the default knowledge base storage quota (default quota is 1000 MB) has not been adjusted, which is insufficient to index multi-source delivery data.
- Retrieval results may include documents unrelated to delivery optimization. The symptom is that irrelevant operational documents or technical manuals appear in the recall results. The cause is that the retrieval scope has not been limited to the dedicated knowledge base collection via
collectionIds, resulting in recall of data from other business domains.
How to Verify Correct Configuration
- Upload a historical price adjustment record document, run a retrieval with the keywords "price adjustment" and "conversion cost", and verify whether the recall results include relevant snippets from this document.
- After enabling
usingReRank, compare the order of retrieval results before and after enabling, to confirm that highly relevant content has a higher priority. - Open the knowledge base management interface, and confirm that the dedicated knowledge base collection for delivery optimization has been specified via
collectionIds. - Test real-time updated landing page conversion data, and confirm that the retrieval results include the latest real-time data snippets.
The values provided 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.