What this category of data looks like
Business data for pulp and paper manufacturing comes from preparation, pulping, papermaking, quality and energy departments. It covers documents such as pulp mixing sheets, beating records, paper machine operation logs, online quality curves, laboratory reports, and sheet break and grade change records. Data sources include wood pulp and waste paper batches, DCS/QCS systems, paper machine online measurement equipment, laboratory systems, and energy and water treatment systems. Update frequencies fall into three categories: real-time, shift-based, and per roll. This category involves high-speed continuous web production, where data links cross-sectional and longitudinal quality distribution and sheet break risk. Most documents include process parameters, batch identifiers, and real-time monitoring fields.
What constraints do these characteristics impose on knowledge base retrieval and recall?
Multiple data sources and update frequencies require differentiated indexing strategies. Real-time monitoring data needs high-frequency index refreshes. Shift-based and per-roll batch documents can be updated incrementally per production cycle. Documents contain a large number of process parameters with clear units, such as grammage, moisture content, and thickness. Retrieval must retain field precision to avoid losing critical information through fuzzy matching. In high-speed continuous production scenarios, incident report data has strict timeliness requirements. Retrieval must prioritize recalling recent relevant records. Cross-sectional and longitudinal quality distribution data has multiple dimensions, so the total number of single-recall results must be limited to avoid context overload that harms conversation fluency.
Configuration Settings
These configurations are based on the knowledge base retrieval module of FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Aligns with the precision requirements of pulp and paper process parameters, avoiding recall of irrelevant historical batch data |
limit | Top 6–8 entries | Controls the context length per conversation, fitting the compact information needs of process retrieval |
searchMode | embedding | Prioritizes matching semantic and numerical features of process parameters and monitoring data, avoiding interference from text fuzzy matching |
usingReRank | Enabled | Re-ranks recall results by process relevance, prioritizing matching current production batch’s pulp mixing and sheet break scenarios |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval terms for process queries, such as expanding "beating parameter adjustment" to field matching for corresponding beating records |
collectionIds | Bind corresponding knowledge base collections by department | Separates exclusive data for preparation, pulping, papermaking and other departments, avoiding cross-department data confusion |
Three Common Configuration Errors
- Symptom: Retrieval results do not match the intended process document content, returning irrelevant historical batch data. Cause: The
similaritythreshold is set too low, recalling non-target data outside the precision range. - Symptom: After configuring a new embedding model, retrieval tests return errors, and the interface displays field mismatch prompts. Cause: The new model was not adapted to the numerical matching logic for pulp and paper process parameter fields (such as grammage, moisture content), leading to index construction failure.
- Symptom: After configuring knowledge base trigger rules, relevant content exists but no knowledge base data is recalled, and a preset answer is returned directly. Cause:
searchModewas not correctly set to the knowledge base matching retrieval mode, or thelimitparameter was set to 0, resulting in no valid results being recalled.
How to Verify Correct Configuration
- Enter the current production batch’s pulp mixing requirement keywords, and check if the recall results include field information from corresponding pulp mixing sheets and beating records.
- Trigger a retrieval related to incident reports, and confirm if the recall results prioritize displaying recent sheet break and grade change records.
- Adjust the
similaritythreshold, and verify if the matching precision of recall results meets the process parameter requirements. - Test a query with no matching content, and confirm if a preset specified answer is returned.
The values provided are common starting points and should be measured against the reader's own samples.