Brand Product Audience Relevance

Brand Product Audience Relevance

Overview

Use this skill to turn a user-provided brand/product into a CID audience relevance analysis. The default output is a 40-row 年龄段 x 城市级别 ranking with two separate objectives: conversion and brand_mind.

Core Rules

• Treat the 40 年龄段 x 城市级别 segments as the primary ranking universe.
• Treat the 30 人生阶段 x 城市级别 segments as auxiliary evidence only.
• Use strict observed CID evidence first: Affinity Index and Chi-Square on observed counts.
• If the product has no direct CID hit but its category is observed, set match_status to category_observed.
• If only proxy variables are available, stop and ask the user before producing inferred rankings.
• Keep conversion and brand_mind separate. Do not collapse them into one score.
• Record external source URL and crawl/read date for brand/product enrichment.
• Normalize product attributes to: 品牌、规格、属性、价格、买点、卖点.
• Explain results with both numeric evidence and plain-language interpretation.

Workflow

1. Standardize the input.

Identify whether the user supplied a brand, product, product family, URL, or free text. If only a product is supplied, enrich the brand from official or public sources.

2. Check CID observability.

Search CID terms for brand, product, product family, category, media entities, and useful profile variables. Use:

• observed for direct brand/product/entity evidence.
• category_observed when the relevant CID category is observable but brand/product is not.
• inferred only after user confirmation.

3. Create an analysis config.

Copy references/config_template.json to a working config path and fill product attributes, sources, features, objectives, and plain-language phrases. Use references/dabao_jinghuashuang_example.json as a concrete example.

4. Run the Excel analyzer.

python3 skills/brand-product-audience-relevance/scripts/analyze_brand_product.py \
  --config path/to/config.json \
  --source-dir "/Users/apple/Downloads/User/CID 人群" \
  --out-dir outputs/brand_product_audience

The script emits:

• *_ranking.csv: 40 primary audiences with conversion and brand-mind ranks.
• *_evidence.csv: feature-level AI, Chi-Square, p-value, and percentile evidence.
• *_report.md: auditable Markdown report.

5. Validate the output.

Confirm:

• primary_segments=40
• auxiliary_segments=30
• ranking CSV has 40 data rows
• report includes sources, product attributes, direct-hit check, methods, Top 10, full 40-row ranking, and boundaries

6. Summarize for the user.

Lead with the result files and top audience patterns. Mention match_status and the biggest data boundary. Keep detailed tables in the generated report/CSV.

Database Utilities

Use PostgreSQL only when reachable. Current default connection:

host=127.0.0.1
port=5444
database=radar
user=radar
password=radar
schema=cid

Check table counts:

python3 skills/brand-product-audience-relevance/scripts/check_database.py

Run a read-only SQL query and output CSV:

python3 skills/brand-product-audience-relevance/scripts/run_readonly_sql.py \
  --sql "select count(*) as n from cid.dim_audience_segment"

Use references/common_queries.sql for reusable query patterns.

References

• references/workflow.md: fuller operating guide and quality rules.
• references/data_contract.md: CID Excel layout, config schema, and output schema.
• references/database.md: database status, target schema extensions, and future brand_product schema.
• references/common_queries.sql: reusable SQL snippets.
• references/config_template.json: blank analysis config.
• references/dabao_jinghuashuang_example.json: completed example config.
• references/db_dumps/radar_cid_70_20260602.dump: current PostgreSQL cid schema migration backup for the 70-table dataset.
• references/db_dumps/radar_cid_30_life_stage_20260602.dump: historical PostgreSQL cid schema migration backup for the original 30 life-stage-city tables.

Load only the specific reference needed for the task.