Knowledge Base Retrieval and Recall for Full-Vehicle Financial Report Analysis

Full-vehicle financial report data primarily comes from publicly disclosed annual reports, quarterly reports, and temporary announcements issued by

What Data for This Category Looks Like

Full-vehicle financial report data primarily comes from publicly disclosed annual reports, quarterly reports, and temporary announcements issued by exchanges. It also draws from official disclosure documents posted to the investor relations sections of automaker official websites.

Quarterly reports release 1 to 2 months after the end of each quarter. Annual reports release within 4 months after the end of the calendar year. Temporary announcements such as sales speed reports and production capacity adjustment announcements release at any time.

Document structures include sections such as financial statements, management's discussion and analysis, vehicle model sales and production capacity data, and ownership structure. Fields cover revenue, per-vehicle gross profit, segmented model sales, and similar metrics. Common units include RMB yuan, units, sets, and other standard units.

Constraints on Knowledge Base Retrieval and Recall

Data sources are scattered and formats vary, including structured reports and unstructured analysis text. Retrieval systems must support multi-source data access and format compatibility.

Update frequency is high, with temporary sudden announcements. Knowledge bases must support flexible incremental update mechanisms to avoid data lag.

Individual documents are lengthy and divided by chapters. Retrieval segmentation strategies must retain chapter-level semantic integrity to prevent semantic breaks from cross-chapter splicing.

Minor differences exist in field units. Retrieval processes must unify unit benchmarks to avoid matching deviations caused by unit mismatches.

Configuration Settings

Configuration ItemRecommended SettingRationale
PARSE_SEGMENT_STRATEGYAutomatically segment by document chaptersAdapts to the chapter-based document structure of full-vehicle financial reports and retains semantic integrity
Recall countTop 8-12 resultsSingle full-vehicle financial report documents have large content volume, requiring sufficient context to support cross-chapter analysis
Similarity threshold0.72-0.80Financial report terminology is highly professional; balances precision and coverage to avoid missing key matches
Incremental Update CycleEvery 7 daysAdapts to the disclosure rhythm of quarterly reports and temporary announcements to ensure data timeliness
UPLOAD_FILE_MAX_SIZE200 MBSingle annual financial report PDF typically ranges from 100-150 MB, reserving upload space
Rerank result countTop 5-6 resultsSecondary filtering of recall results to focus on highly relevant content and reduce context redundancy

The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require case-by-case analysis. Testing on local samples is recommended before finalizing settings.

Three Common Configuration Mistakes

  • Phenomenon: When creating a knowledge base, the text understanding model dropdown list is empty, and the corresponding model cannot be selected. Cause: A valid API key for the corresponding large model has not been configured in the platform backend, or the key does not have project access permissions.
  • Phenomenon: Retrieval results only return 1 matching text block, which cannot support complete full-vehicle financial report analysis. Cause: The Recall count parameter has not been adjusted, and the default single-recall configuration is used, which does not adapt to the long-text analysis requirements of full-vehicle financial reports.
  • Phenomenon: An upload failure error triggers when uploading a single annual financial report PDF. Cause: UPLOAD_FILE_MAX_SIZE has not been adjusted to above 200 MB, and the default configuration limits large file uploads.

How to Verify Successful Configuration

  • Enter the configuration interface of the target knowledge base, check the value of PARSE_SEGMENT_STRATEGY, and confirm that it matches the setting for financial report chapter segmentation.
  • Submit a test query containing core terminology for full-vehicle financial reports, and check whether the number of returned retrieval results matches the configured Recall count.
  • View the incremental update logs of the knowledge base, confirm that automaker financial report announcements disclosed in the past 7 days have been automatically synchronized, and verify that the update cycle setting is effective.
  • Open a single parsed financial report document, check whether the segmented text blocks retain chapter semantic integrity, and no cross-chapter truncation has occurred.

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.