What the data for this category looks like
Usually, core data for this category comes from search ad backends, search term reports, bidding ranking data, and landing page data. Update frequency is daily or weekly. Commonly used documents are mostly structured tables. These include keyword lists, negative keyword lists, bid records, and search term reports. Fields include keyword text, match type, single bid, click count, conversion count, and more. Units include basic ones like yuan per click, clicks, and count. Some documents have formatting details such as merged cells and blank lines.
Constraints Imposed by These Characteristics on Document Parsing and Chunking
The mostly structured table document structure requires the parsing step to retain row-column associations. It must avoid splitting associated fields like keywords and their corresponding bids or matching rules. The daily or weekly update frequency creates a need for batch uploads. The system must adapt to time requirements for parsing multiple documents simultaneously. Fields contain mixed text and numerical types. Chunking must ensure the integrity of grouped data, preventing loss of associated relationships across chunks. Some original documents have inconsistent formatting. The system must support parsing non-standard formats like merged cells and blank lines.
Configuration Settings
These configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Documents in this category are mostly tightly linked structured tables. Each chunk must contain a complete keyword group and its corresponding fields such as bid and match type, to avoid splitting associated data |
chunkSplitter | Split by table row groups | Structured table data must retain the binding relationship between each row's keywords and corresponding fields, to avoid data association breaks caused by cross-row splitting |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | When batch uploading documents such as daily or weekly search term reports, sufficient parsing time must be reserved to avoid task interruptions |
chunkSettingMode | Manual configuration | Fields for documents in this category are fixed, and the association logic is clear. Custom chunking rules can better adapt to business data characteristics |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapt to the need for batch uploading large-volume tabular documents such as weekly search term reports, and meet the capacity requirements for batch processing |
Three Common Configuration Mistakes
- After uploading a knowledge base generated from a Feishu online spreadsheet, the exported dataset.csv only contains template formatting and no actual keyword data. The cause is that parsing fails to correctly identify the nested cell structure of the online spreadsheet, leading to failed field mapping.
- After chunking, a large number of small chunks containing only a single keyword appear, with no corresponding bid or match type information. The cause is that the chunk splitting rule for splitting by table row groups is not enabled, and general text splitting logic is mistakenly applied to structured table data.
- When batch uploading multiple weekly search term reports, parsing tasks frequently time out. The cause is that the
PARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted upward, failing to adapt to the parsing time required for batch large-volume documents.
How to Confirm Correct Configuration
- Upload a single tabular document containing 100 rows of keywords. Check the parsed chunk results. Confirm that each chunk contains at least one complete set of keywords and their corresponding fields such as bid and match type.
- View the parsing task logs. Confirm that no timeout errors occur, and task execution time falls within the interval set by the
PARSE_FILE_TIMEOUT_SECONDSparameter. - Export the chunked dataset of the knowledge base. Confirm that all fields are fully populated, with no empty fields or template placeholder content.
- Upload multiple weekly batch documents. Check the success rate of batch parsing. Confirm that no batch task interruptions occur.
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.