What the data for this category looks like
Data for chemical pharmaceutical financing daily reports comes from three main sources: listed company announcements on the Shanghai Stock Exchange, Shenzhen Stock Exchange, and Beijing Stock Exchange; financing information for unlisted chemical pharmaceutical enterprises disclosed by local equity trading centers; and third-party pharmaceutical industry databases. Documents are updated every workday. Each daily document contains dozens of financing entries. Documents use structured tables as the core carrier, supplemented by short event description text. Included fields are: financing entity name, affiliated chemical pharmaceutical sub-sector (such as chemical raw materials, chemical preparations), financing round, financing amount (unit: ten thousand yuan or hundred million yuan), investor list, disclosure date, disclosure source, and others.
What constraints do these characteristics impose on document parsing and chunking?
The high proportion of structured tables requires the parsing process to accurately identify merged cells and the continuity of cross-page tables. Avoid splitting table rows, which causes field misalignment. Financing amounts use two units: ten thousand yuan and hundred million yuan. During chunking, retain the binding relationship between fields and units to prevent mismatched amount values and units. Daily documents contain multiple independent financing entries. Chunk by single financing event as the minimum unit to avoid context confusion caused by cross-entry splicing. Some unlisted enterprises have inconsistent financing information formats. Adapt to multiple table layouts while fully retaining the association between the investor list and financing entity.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
PARSE_TABLE_ENABLE | Enabled | Documents use structured tables as the core carrier. Enabling this option accurately extracts table fields and content |
TABLE_CHUNK_SPLIT_MODE | Split by row groups, with each group corresponding to a single financing event | Daily documents contain multiple independent financing entries. Splitting by row groups ensures each chunk corresponds to a complete financing event |
PARSE_UNIT_AWARE | Enabled | Financing amounts use two units (ten thousand yuan and hundred million yuan). Enabling this option retains the binding relationship between fields and units |
MAX_CHUNK_SIZE | 800–1200 characters | Text and table content for a single financing event typically falls within this range, ensuring complete chunk context |
PARSE_FILE_TIMEOUT_SECONDS | 120 seconds | Daily documents have a large number of entries. Reserve sufficient time for table recognition and chunking during parsing |
RETAIN_TABLE_STRUCTURE | Enabled | The association between table fields and content is critical for subsequent retrieval. Enabling this option retains the original table layout and field correspondence |
The parameter values provided on this page are common starting points for configuration. Actual values are affected by material format, data volume and business rules. Specific issues require individual analysis, and it is recommended to test on your own samples before finalizing.
Three Common Configuration Mistakes
- Phenomenon: Chunks stored in the knowledge base contain duplicate financing events, and the index order after custom chunking does not match the original document. Cause: Did not use single financing event as the minimum unit for chunking, mistakenly used the full page table or document as the chunk unit, leading to the same financing information being split into multiple chunks or cross-entry splicing, causing duplicate storage and order confusion.
- Phenomenon: Parsed chunks have messy table content formatting, cannot be presented in the original table structure, and some cell content is split into different chunks. Cause: Did not configure
TABLE_CHUNK_SPLIT_MODEto split by row groups, leading to incorrect splitting of table rows and destroying the integrity of single financing events. - Phenomenon: When creating a knowledge base via API, real-time status of the parsing task cannot be obtained, and it is impossible to determine whether parsing is completed or failed. Cause: Did not enable the status return parameter in the API request, or did not configure the synchronous reporting logic for parsing status, resulting in inability to obtain status information such as parsing in progress, ready, or failed.
How to Confirm Correct Configuration
- Upload a single test document containing 2-3 different financing events, view the parsed chunk list, confirm each chunk corresponds to a complete single financing event with no cross-entry splicing.
- Randomly select the financing amount field from the chunks, check that the amount value and corresponding unit are bound together, with no separation or misalignment.
- Call the API to initiate a parsing task, check if the returned response parameters include the parsing status field, confirm that status information such as parsing in progress, ready, and failed can be obtained.
- View the parsed chunk content, confirm that the table structure is fully retained, and cell content is not incorrectly split or misaligned across rows.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.