What the data for this category looks like
Financial report data for listed consumer building materials companies comes from two main sources: official disclosure platforms of the Shanghai Stock Exchange and Shenzhen Stock Exchange, and official annual, semi-annual, quarterly reports and temporary announcements published by listed companies. Update schedules align with disclosure cycles. Annual reports are released once per year. Semi-annual and quarterly reports are updated quarterly. Temporary announcements are released immediately for major operating events. Most documents use PDF format, with some including docx format meeting minutes attachments. Documents contain consolidated balance sheets, income statements, cash flow statements, corresponding statement notes and management discussion and analysis content. Fields include current revenue, attributable net profit, net operating cash flow and other financial indicators, with units marked in ten thousand yuan or hundred million yuan.
What constraints these characteristics impose on knowledge base retrieval and recall
Data sources for consumer building materials financial reports are concentrated on official disclosure platforms. Updates must align with quarterly, annual and temporary event disclosure rhythms. This requires the retrieval and recall link to support precise filtering of results by disclosure time. This prevents the return of outdated data. Most financial report documents are long-form text, with nested financial tables and notes. This requires the parsing link to accurately extract financial fields from tables. This avoids missing key data such as revenue and costs of segmented categories. Financial indicators use different units such as ten thousand yuan and hundred million yuan. Retrieval must match unit dimensions. This prevents unit confusion in recalled results. Temporary announcements are released irregularly. This requires incremental synchronization mechanisms to adapt to non-periodic content updates. This ensures the timeliness of recalled data.
How to set the configuration
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 8000–12000 characters | Single consumer building materials financial report document can be dozens of pages long. A long context window is required to fully accommodate recalled financial report fragments and queries. |
similarity threshold | 0.72–0.8 | Financial report fields are mostly structured numerical values. A higher threshold is needed to filter low-match irrelevant documents, avoiding recall of non-target financial report content. |
Maximum number of knowledge base citations | Top 3–5 entries | Key financial fields of a single financial report are concentrated. Too many citations will exceed context limits, leading to truncation of core information. |
PARSE_PDF_TABLE | Enabled | Consumer building materials financial reports contain a large amount of tabular data for segmented category revenue and costs. PDF table parsing must be enabled to accurately extract fields. |
UPLOAD_FILE_MAX_SIZE | 500 MB | Annual financial report PDF files are usually large in size. The large file upload limit must be adapted. |
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | Long document parsing takes a long time. This prevents parsing failure due to timeout. |
The parameter values provided on this page are all conventional recommendations, used as a starting point for configuration. Actual values are affected by material form, data volume and business rules. Specific issues require specific analysis. It is recommended to test on your own samples before finalizing.
Three common errors
- Phenomenon: The answer generated by the test workflow always attaches system prompt text not present in the knowledge base at its end. Cause: The default brand introduction content is not disabled in the system prompt, leading the model to add it automatically during generation.
- Phenomenon: Docx format financial report attachments imported into the knowledge base cannot extract embedded financial table data. Cause: Table recognition configuration in document parsing is not enabled, leading to incorrect parsing of table content within docx files.
- Phenomenon: The number of knowledge base recall results exceeds the context limit, and the answer is forcibly truncated. Cause: The
Maximum number of knowledge base citationsparameter is not set reasonably, introducing too many non-core financial report fragments.
How to confirm the configuration is correct
- Upload a single complete annual financial report PDF of a consumer building materials listed company. Check the extraction results in the document parsing list to confirm that no fields are missing from the segmented category revenue table.
- Initiate a query containing specific financial fields and time ranges. Verify that the source of the recalled knowledge base fragments is the corresponding chapter of the target financial report.
- Adjust the
similarity thresholdparameter to verify that the match degree of recalled results meets business screening requirements. - Import docx format financial report meeting minutes attachments. Confirm that embedded text and table content are correctly parsed and included in the knowledge base.
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.