What This Category of Data Looks Like
The business data for search ad keyword management comes primarily from search ad backends, search term reports, bidding ranking data, and landing page data. Most updates occur daily or weekly. Daily generated documents include structured keyword lists, negative keyword tables, bid records, and search term reports. Fields include keyword text, matching mode, bid threshold, historical search volume, conversion cost, and more. Units include yuan, search counts, thousand impression counts, and others. Business stakeholders are SEM optimizers, agency operations staff, and brand marketing personnel. Data is generated across the full ad delivery lifecycle tied to user-initiated searches, with no additional cross-category associated data.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall?
Structured ad business data requires retrieval to precisely match fields such as keywords and matching modes. This avoids irrelevant content interfering with on-call operations and ticket responses. Daily or weekly high-frequency updates require the knowledge base to support incremental synchronization. This ensures retrieval results always correspond to the latest delivery data. The rapid response requirement in ticket scenarios limits the number of results and context length returned per retrieval. This avoids delaying on-call operation processing timelines. Additionally, the strong association between negative keyword tables and keyword lists requires retrieval to support cross-document associated matching. This improves problem-solving efficiency. Consequences of errors such as invalid clicks or brand term hijacking also require retrieval recall accuracy to remain at a high level. This avoids delivering incorrect ad delivery rules.
How to Configure Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.88 | Matches text similarity for search ad keywords, balances precise recall and coverage, and adapts to structured keyword data characteristics |
limit | Top 12–18 entries | Controls the number of results returned per retrieval, adapts to fast response needs of operation and maintenance ticket scenarios, and avoids excessive content increasing context processing costs |
usingReRank | Enabled | Performs re-ranking on initial recall results, improves display priority for highly relevant keywords and negative keyword records, and adapts to strong correlation requirements of ad data |
datasetSearchUsingExtensionQuery | Enabled | Expands semantic scope of user queries, covers vague ticket questions such as "last week's negative keyword table" matching documents updated in the corresponding weekly cycle |
collectionIds | Bind IDs of corresponding business datasets | Isolates exclusive data such as search ad keyword lists and negative keyword tables, avoids retrieving irrelevant documents from other businesses |
Three Common Configuration Mistakes
- Symptom: Retrieval returns empty results, but normal results are returned during internal knowledge base testing. Cause:
collectionIdsis not correctly configured, and the bound dataset does not include exclusive documents for search ad keyword management, leading the retrieval scope to deviate from business data. - Symptom: The knowledge base has no updated content for an extended period, and the latest keyword or negative keyword data cannot be obtained. Cause: Data sources from search ad backends and search term reports are not configured for daily or weekly synchronization, leading knowledge base data to lag behind business update cycles.
- Symptom: After upgrading FastGPT, clicking the knowledge base pops up an error prompt, and the knowledge base function cannot be accessed normally. Cause: Retrieval configuration parameters from older versions are not compatible with the new version's interface specifications, and logs are not viewed through the console to locate specific errors.
How to Confirm Correct Configuration
- Manually enter a known existing keyword, check if retrieval results include the corresponding keyword list and negative keyword table documents, and confirm the retrieval scope covers business data.
- After enabling
usingReRank, check if retrieval results prioritize displaying keywords and data related to high bids and high conversions, to verify the re-ranking function is active. - View dataset synchronization logs for FastGPT v4.17.0, confirm that daily or weekly ad backend data has been successfully synchronized to the knowledge base.
- Check that
collectionIdsonly includes business datasets for search ad keyword management, to avoid mixing in irrelevant documents from other businesses.
The configuration values listed above are common starting points. They should be evaluated against the samples of the relevant use case.
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.