Preface

Customer feedback for chain restaurants is scattered across five channels: brand surveys, Dianping, Meituan, Taobao Flash Delivery, and JD Instant Delivery. Each channel has different fields, rating scales, and export formats. Manually consolidating one round typically takes half a day, and the result is often just a pile of text that fails to translate into actionable conclusions at the store or regional level.

The Skill described below turns this workflow into a fixed pipeline: batch import → field standardization → QSC scoring → NPS calculation → NLP sentiment and root cause analysis → regional/store comparison → HTML dashboard → weekly/monthly reports → Excel issue list. No self-hosted services are required; the scripts come with a --demo mode.

What This Is

In one sentence: a multi-platform customer feedback analysis tool for chain restaurants, outputting QSC three-dimensional scores, NPS, sentiment and root cause tags, as well as ready-to-distribute dashboards and reports.

  • Skill ID: @user_483ab805/chain-store-customer-feedback-analysis
  • Version: 1.0.0
  • Source: community (community-published, not an official team product)
  • Category: data-analysis
  • Tag: requires_api_key=false (the default local NLP path does not need an API key)
  • Publisher: user_483ab805

It solves this problem: applying the same scoring methodology across all feedback channels, and breaking down results along two dimensions—store and time—eliminating the repetitive manual consolidation work each period.

Core Capabilities

The entire pipeline consists of 7 layers, described below layer by layer.

1. Data Layer: Batch Import from 5 Platforms

Supports batch import from five sources: brand surveys, Dianping, Meituan, Taobao Flash Delivery, and JD Instant Delivery. After import, all sources enter the same processing pipeline, eliminating the need to maintain separate scripts per platform.

2. Field Standardization

Unifies the schema by mapping different platforms’ field names, rating granularities, and text formats into a single structure. Simultaneously performs customer ID desensitization, ensuring no identifiable customer information is retained in exports or downstream analysis.

3. Scoring Layer: QSC Three-Dimensional Scoring

Scores each feedback item across three dimensions: Quality (Q), Service (S), and Cleanliness/Environment (C). The three dimensions are calculated separately to help pinpoint whether issues lie in product or on-site management.

4. NPS Calculation

Calculates NPS using the promoters-minus-detractors methodology, distinguishing between “willingness to recommend” and “satisfaction” as two different questions.

5. NLP Layer: Sentiment and Root Cause

Defaults to a local approach: jieba tokenization + Chinese sentiment dictionary (2000+ words) + 10 root cause tags. This path requires no API key and keeps data local.

Optional upgrade to cloud NLP: Alibaba Cloud NLP or Baidu Cloud NLP, with users providing their own keys. The default tag set is fixed at 10 major root causes, covering common issue categories in restaurant feedback.

6. Alerting and Regional Comparison Layer

Issues alerts for negative sentiment. Regional/store comparison cross-references two dimensions: store + time, so it can simultaneously answer “which store is performing poorly” and “is this store getting worse.”

7. Dashboard and Reporting Layer

  • HTML dashboard: 6 Plotly charts in a single file.
  • Weekly/monthly reports: HTML single files for email or internal circulation.
  • Excel issue list: aggregated by store, for manual distribution to stores. Does not integrate with ticketing systems; if you need a store remediation closed-loop, you must build that yourself.

Trigger Words

The Skill uses trigger words to determine which branch to follow. The word list is as follows:

  • Feedback aggregation: customer feedback / post-consumption survey / review / bad review / bad review analysis / good review analysis / sentiment analysis / public sentiment / word-of-mouth / evaluation / comment / review
  • QSC scoring: QSC / quality / service / cleanliness/environment
  • NPS calculation: NPS / net promoter score / willingness to recommend
  • Report generation: weekly report / monthly report / data dashboard / regional comparison / regional ranking / store ranking / brand public sentiment
  • Multi-platform import: Dianping / Meituan / Taobao Flash Delivery / JD Instant Delivery / brand survey
  • NLP analysis: sentiment analysis / semantic analysis / negative sentiment / bad review alerts
  • Cloud NLP: Alibaba Cloud NLP / Baidu Cloud NLP / API key

When a trigger word is matched, the corresponding stage is entered. The entire flow remains a 7-layer pipeline, not seven independent workflows.

Installation and Activation

It should be noted that at the time of writing, the official pages (api.skillhub.cn and skillhub.cn) were both inaccessible, so installation commands and directory structures could not be verified. The content below only covers what can be confirmed.

  • This Skill is based on the general SKILL.md format. To activate it, place it as a skill in the client’s skills directory. The specific path depends on the client’s documentation.
  • requires_api_key=false: the default local jieba + sentiment dictionary path works out of the box without preparing a key.
  • If switching to cloud NLP, you need your own Alibaba Cloud NLP or Baidu Cloud NLP key, and you are responsible for the associated call costs.
  • For data sources, you need access to the raw export files from each of the five platforms.

Typical Usage

All pipeline scripts support --demo mode to run through the full workflow, allowing you to verify the pipeline before connecting real data:

# Add --demo to each layer's script to run the entire pipeline with built-in sample data
# Script names are subject to the actual repository (the official page is currently inaccessible, so they could not be verified)
python <data>.py   --demo   # Import and standardization
python <score>.py  --demo   # QSC and NPS
python <nlp>.py    --demo   # Sentiment and root cause tags
python <report>.py --demo   # Dashboard and weekly/monthly reports

The --demo mode skips real data and uses sample data to walk through all 7 layers, making it suitable for confirming during initial deployment whether the environment and outputs meet expectations.

In real usage, the inputs are raw feedback files from the five platforms. The outputs are: standardized detail records, QSC/NPS scoring results, feedback lists with sentiment and root cause tags, regional and store comparisons, HTML dashboard, weekly or monthly report HTML single file, and Excel issue list aggregated by store.

The final step requires manual intervention: after the Excel issue list is aggregated by store, you distribute it to the corresponding stores. The Skill itself does not handle dispatching tasks.

Use Cases and Caveats

Suitable for these situations:

  • Multi-store chain restaurants with feedback from more than two platforms, currently relying on manual consolidation.
  • Need for unified QSC and NPS standards, rather than each platform having its own calculation method.
  • Headquarters or regional operations needing periodic dashboards and weekly/monthly reports.
  • Sensitive about data leaving the premises, preferring sentiment analysis to be done locally.

Points to note:

  • The default sentiment analysis is a dictionary-based approach (2000+ words), which has limited recall for sarcasm and long compound reviews. For higher accuracy, switch to cloud NLP, but you must provide your own key and bear the costs.
  • Does not integrate with ticketing systems; distribution of the issue list and remediation closed-loop must be handled by you.
  • Version 1.0.0, community source, 1164 downloads, 0 installations. The zero installation count means there are no existing deployment feedback references. Before connecting production data, it is recommended to run --demo first and spot-check the accuracy of QSC and sentiment tags.
  • The score displayed as 100000 is a platform-side display value and is not recommended as a basis for quality assessment.
  • The official page is currently inaccessible. Installation commands, directory structure, and script naming are subject to the actual repository contents.

Conclusion

The value of this Skill lies in transforming the repetitive and error-prone task of “multi-platform feedback consolidation” into a fixed 7-layer pipeline with unified standards, reproducibility, and report generation. The default local NLP path has no key barrier, and --demo can fully verify the pipeline.

Official URL: https://api.skillhub.cn/user_483ab805/chain-store-customer-feedback-analysis