City Public Opinion Monitor

What Problem It Addresses

City managers, government teams, and public-affairs researchers often have to chase scattered signals across news sites, social platforms, forums, and government channels. Manual review is slow, and sentiment judgments are hard to audit. city-public-opinion-monitor turns that raw text into traceable monitoring output: which topics are rising, whether negative sentiment is abnormal, what risk level is indicated, and whether report claims can be traced back to source data.

How It Works

The skill follows a six-step workflow:
- Multi-source collection: gathers relevant content using keywords, geographic scope, and time windows.
- Sentiment analysis: labels text as positive, neutral, negative, or sensitive, with judgment evidence and switchable lexicon, ml, or llm strategies.
- Topic clustering: uses TF-IDF, TextRank, LDA, and co-occurrence analysis to surface sub-issues.
- Risk identification: applies weighted signals such as propagation heat, negative share, acceleration, influential-account participation, and cross-platform spread.
- Trend prediction: estimates momentum, emotional evolution, and possible escalation points using propagation models, sentiment curves, and event-calendar context.
- Report generation: produces daily, weekly, special, monthly, or quarterly reports with data sources, collection time, generation timestamp, and AI-assistance disclosure.

Boundaries and Caveats

It is suited to city governance, policy feedback, public incidents, and major-event monitoring. The source material explicitly excludes entertainment marketing, competitive negative digging, post removal or review manipulation, and pure NLP model training. When data is empty, scarce, or polarity is one-sided, statistical conclusions should be downgraded. Claims must remain traceable to original text or the source list; inference should not be presented as sourced fact.