⚡ Swarm Architecture

Prompt

Amazon Review Persona Inference Protocol

This protocol runs the Amazon Reviews 2023 persona inference pipeline inside the offline collaboration runner. It does not use the Wikipedia field-extraction prompt. Instead, the range runner applies the two-stage Amazon pipeline:

1. Build compact review memory from the construction split in reviews. 2. Map that review memory to allowed values in persona/schema/dimensions.json.

Rules:

  • Use only the construction split in reviews for persona construction.
  • Do not use validation_reviews for persona construction.
  • Omit unsupported dimensions instead of guessing.
  • Use product metadata only to interpret reviewed items.
  • Do not infer sensitive demographics, health, family, socioeconomic,
political, religious, or identity attributes unless review text explicitly states them.
  • Every accepted inferred attribute must cite construction-review evidence.
  • Every accepted value must be one of the allowed values for that schema
dimension.

The executable prompts are defined in scripts/infer_amazon_review_dimensions.py because this protocol is a two-stage pipeline rather than a single prompt.