What the Data for This Category Looks Like
This knowledge base supports white goods marketing customer acquisition scenarios. The platform sources data primarily from brand-official product specifications, marketing copy, and installation and maintenance manuals. The platform adjusts update schedules based on new product launches and promotional activities, with no fixed cycle. It pushes full parameter content updates in bulk when new products launch. It temporarily adjusts promotional-related copy and price information before promotional activities. The platform splits and categorizes documents by product model. Each document includes structured parameter fields and semi-structured marketing copy. Parameter fields cover model, energy efficiency rating, external dimensions, rated power, and recommended retail price, with corresponding units: none, rating, millimeters, watts, yuan. Marketing copy includes scenario-based usage suggestions, promotional activity explanations, and frequently asked questions for users. The platform primarily uses PDF and Word formats for documents.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall
The platform splits documents by model and includes structured parameters and marketing copy. Retrieval processes must accurately match the appliance model mentioned by users to avoid recalling irrelevant cross-model content. Parameter fields have clear units. Retrieval processes must retain unit associations to prevent parameter confusion. The platform updates marketing content temporarily for promotional activities. It must trigger regular vector database reindexing to ensure the timeliness of recalled content. Marketing copy varies significantly across different models. Retrieval processes must group recalls by model dimension to improve matching accuracy. Some documents include scenario-based usage suggestions. Retrieval processes must distinguish between product parameters and scenario content to avoid recalled content deviating from marketing customer acquisition goals. White goods marketing materials often include promotion deadline labels. Retrieval processes must filter recalled content to remove expired promotional copy, ensuring output content is compliant and effective.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
Recall count | Top 6-8 entries | White goods have relatively abundant marketing content per model. Too many recalled entries will exceed the context window. Too few will fail to cover complete parameters and copy. |
Similarity threshold | 0.72-0.78 | This range distinguishes precise model parameter matching from generalized marketing scenario matching, and filters low-relevance cross-model content. |
Chunk size | 800-1200 characters | White goods documents include parameter blocks and long-form marketing copy. This segment length preserves parameter integrity and copy coherence. |
Rerank result count | Top 3-4 entries | Focuses on core marketing content and key parameters, avoiding redundant information interfering with generator output. |
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | Some product manual documents have large length, requiring sufficient parsing time reserved. |
maxContext | 8000 characters | Adapts to context splicing requirements for multi-model parameters and marketing copy, avoiding truncation of core information. |
The parameter values provided on this page are common starting points for configuration setup. Actual values are affected by material format, data volume and business rules. Each scenario requires targeted analysis. It is recommended to test on your own samples before finalizing values.
Three Common Mistakes
- Phenomenon: When calling the
/api/v1/chat/completionsinterface, no knowledge base uploaded content is retrieved after specifying the appId. Causes: The target knowledge base is not associated with the corresponding appId, the correct knowledge base binding parameter is not included in the interface request, or the knowledge base has not completed full vector index construction. - Phenomenon: The generated answer does not include source document information for the retrieved passages. Causes: The traceability display configuration of the knowledge base is not enabled, or the output format of traceability content is not configured in the system prompt.
- Phenomenon: Retrieved content only includes text, without image information from the document. Causes: The image vectorization and associated storage settings during document parsing are not enabled, or the generation prompt does not explicitly require referencing image content.
How to Confirm Configuration is Complete
- Upload a single white goods product manual and marketing copy, review the parsed text segments and field extraction results, confirm that models, parameters and marketing copy are correctly split and categorized.
- Initiate a query that includes specific appliance models and parameter requirements, check the matching degree and quantity of recalled results, adjust configuration items to values suitable for the business scenario.
- Enable the traceability display configuration, initiate the same query, confirm that the answer includes source document identifiers for retrieved passages.
- Call the interface for the specified appId, verify that the returned results include white goods-related content from the knowledge base, confirm that the association relationship is active.
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.