Marketplace promotions: multilingual work with vector indexes

Review vector indexes for multilingual work in Marketplace promotions. Check document scope, failure paths and source evidence through controlled tests.

What the documents look like

For multilingual work in Marketplace promotions, this page examines FastGPT vector indexes. Distinguish proposed submission, pending review and selection across language entries for one promotion. The examples are hypothetical.

Constraints at the vector indexes step

A hypothetical promotion note proposes submitting a multipack for the current event and leaves the separate-item campaign to be checked. Its English counterpart preserves that submission plan. An older event letter lists a selected multipack for a different occasion. For an English question about products already selected this time, proposed submission, pending review and selection remain different stages. Inspect whether the entry changes planned submission to selected and whether it points to the current note or the older letter. A matching product name does not establish a matching decision. Event names also need the subject and period stated in the original before they can identify the applicable occasion.

Configuration and implementation checks

Promotion-language checkProposed materialSpecific inspection
Submission stageMultipack proposed for this eventPreserve the plan without adding selection.
Product formMultipack and separate itemDo not transfer the former’s scope to the latter.
Event identityCurrent note versus another event letterCheck subject and stated applicable period.
Decision evidenceCurrent selection notice absentAn index hit cannot replace the notice.
Added-source retestA notice is supplied laterRepeat the question and trace applicable items to it.

Documented conditions

Documented item / implementation checkCondition
Embedding modelKnowledge bases associated with one search node must use the same embedding model.
Original data and indexesOne original data block can have multiple indexes.
Index contentIndex wording should be complete and focused.
Image indexesWhere image retrieval is used, inspect descriptions or image vectors.
Version and result checkAfter a change, reuse questions to compare originals.

Three common errors

  • If an English index adds selection to a submission plan, compare the note with the proposed entry and identify the missing decision notice.
  • If a current-event question hits an older letter, inspect event identity and listed product form. The same multipack name does not transfer an event decision.
  • If the separately checked single-item scope disappears from the actual citation, compare both limiting passages with the answer’s list to see whether multipack eligibility was extended.

How to verify the setup

Ask in both languages about the submission plan, review result and single-item scope, tracing each answer to the passage actually retrieved. When a current notice arrives, check product form and applicable period before repeating the questions. Several indexes add search entrances without supplying a missing selection notice. Where one search node associates several knowledge bases, inspect their actual embedding models and then the event identities used by retrieval and citation. This exercise submits no registration and establishes neither campaign qualification nor an effective discount.

Material contextScope
Document typesActivity Registration Form, Promotional Price Approval Form, Inventory Lock Sheet and Activity Review Form
Candidate sourcesTmall Marketing Backend, JD Activity Backend, ERP Inventory and Historical Transaction Prices
Technical baselineFastGPT v4.17.2