Sentiment Compass

Problem

Chinese social media sentiment is often scattered across Xiaohongshu, Douyin, Weibo, and WeChat Official Accounts. Manual keyword checks, sentiment judgment, and report building are slow, and negative spikes can be discovered too late. Sentiment Compass turns public posts from these platforms into queryable data and uses GLM-4 for Chinese-language sentiment classification.

How It Works

The skill uses Playwright headless to crawl public search results, with UA rotation, random 3-8s delays, and retries to reduce blocking risk. Data is stored in local SQLite and JSON configuration. Core workflow:

• Create a keyword monitoring task and run the first crawl
• Batch-classify posts as Positive, Neutral, or Negative with reason summaries via GLM-4
• Generate a report with total mentions, sentiment ratio, trending charts, and top posts
• Push alerts through a Feishu webhook or SMTP email when a negative threshold is exceeded, and schedule runs with OpenClaw Cron at 6h, 12h, daily, or weekly intervals

Boundaries

It only supports public pages and does not cover login-gated content. Sentiment accuracy depends on text length and context. Tier plans may limit keywords, platforms, crawl frequency, and data retention. It is useful for lightweight brand or competitor sentiment monitoring, not as a final compliance-audit or enterprise-scale precision analytics system.