Dianpingbao AI Review Analytics

Problem

Restaurant owners often have a pile of Dianping or Meituan reviews, but the signal is noisy. Platform star ratings can be inflated, fake reviews can mask real weaknesses, and a single score rarely tells a single-store owner whether the issue is product, service, environment, or value. For chains, the harder question is whether multiple locations are consistent and which location is underperforming. This skill turns that diagnostic into a repeatable workflow: upload reviews you have rights to analyze and get a structured report instead of a vague rating.

How it works

It accepts CSV, Excel, or pasted review text, and focuses on fields such as review content, rating, time, and store. The main steps are:

• Authenticity checks: flag templated wording, ultra-short high-star reviews, and suspicious bursts, then exclude them from scoring while disclosing the count.
• QSCV semantic tagging: score product, service, environment, and value dimensions by sentiment and strength.
• Composite scoring: apply default or category-specific thresholds to produce grades, strengths, weaknesses, and prioritized fixes.
• Multi-store aggregation: when the store field contains multiple values, generate store rankings and cross-store consistency notes.

It can run in LLM semantic mode for nuance such as sarcasm, or in an offline dictionary mode for quick structural checks without external dependencies.

Boundaries

For delivery-grade analysis, use an exported merchant backend review base table with hundreds or thousands of rows. If there are fewer than 15 valid reviews, treat the result as low confidence. The skill does not scrape platforms, does not generate PDF, PPTX, or image files, and may need openpyxl for .xlsx; otherwise, export to CSV.