What the Data for This Category Looks Like
The data sources for the fertilizer and pesticide category mainly include registration announcements released by the National Institute for the Control of Agrochemicals, fertilizer product standard texts released by the Ministry of Agriculture and Rural Affairs, internal enterprise production and quality inspection records, and field test reports. The data update rhythm is adjusted according to policy requirements and product iterations. Standard texts are updated every 1 to 2 years. Internal enterprise production records are synchronized daily. Document structures are divided into three categories: compliance documents (registration certificate application forms, compliance inspection reports), production documents (feeding lists, quality inspection sheets), and technical documents (formulation instructions, field guidance plans). Fields include active ingredient content, dosage form, registration certificate number, and applicable crops. Units mostly use professional measurement standards such as g/L, kg/ha, and %.
Constraints Imposed by These Characteristics on Knowledge Base Retrieval and Recall
Compliance documents are mostly thousands of words of formal text, with fixed-format approval fields. Retrieval must prioritize matching precise fields such as registration certificate numbers and active ingredients to avoid generalized recall of irrelevant content. Diverse professional measurement units require automatic normalization processing during retrieval to prevent retrieval failures caused by differences in unit expressions. Large format differences exist between internal enterprise production records and official standard texts. Different retrieval matching rules must be set according to dataset classification. A high proportion of long texts requires adjustments to the number of recalled entries and chunking logic to prevent key compliance information from being truncated.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Fertilizer and pesticide documents contain a large number of professional terms and precise fields. A threshold that is too low will introduce irrelevant recalls, while a threshold that is too high will miss relevant compliance documents |
limit | Top 8–12 entries | Single compliance documents have long content. Too many recalls will exceed the context window, while too few will fail to cover complete compliance requirements |
searchMode | semantic | Balances professional term matching and semantic understanding, adapting to mixed retrieval scenarios of technical and compliance documents |
usingReRank | Enabled | A large number of long texts and professional fields are present. Reranking filters low-relevance results and improves the accuracy of final recalls |
datasetSearchUsingExtensionQuery | Enabled | For precise fields such as registration certificate numbers and active ingredients, synonymous queries are automatically expanded to avoid retrieval failures caused by expression differences |
collectionIds | Specified by dataset classification | Official standard libraries and internal enterprise libraries are distinguished, and matching rules are set separately to ensure reasonable retrieval weights for documents from different sources |
Three Common Misconfigurations
- Phenomenon: The number of returned results during retrieval does not meet expectations, or the recalled content does not match professional queries. Cause: The
similaritythreshold is not adjusted for fertilizer and pesticide documents' precise fields, and unreasonable threshold settings filter valid content or introduce irrelevant results. - Phenomenon: Image content in the knowledge base cannot be output during conversations, only text descriptions are displayed. Cause: The image analysis and embedding process is not configured, or the image URL is not correctly associated with the knowledge base document, resulting in only text content being extracted during retrieval.
- Phenomenon: After importing PPT-format fertilizer and pesticide documents, only text is extracted without images. Cause: The image analysis switch for PPT documents is not enabled, or floating layer images are not correctly identified and associated with text blocks.
How to Confirm Proper Configuration
- A standard pesticide registration certificate document is uploaded, the query "The active ingredient content of this product" is entered, and the matching score of the returned results is checked to confirm it falls within the 0.75–0.85 range.
- A fertilizer product instruction manual containing images is imported, the query "View the usage method of this product" is launched, and correct display of associated image content in conversations is confirmed.
- A PPT-format production record document is imported, and verification that both text and image content exist in the knowledge base without loss or omission is completed.
- A query containing duplicate registration certificate numbers is launched, and automatic merging of retrieval results into a single citation without duplicate entries is confirmed.
The values provided here are common starting points and should be measured 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 2026-09-28.