Sentiment Analysis Monitor

AI-powered social media sentiment monitoring and analysis tool for Chinese platforms. Monitor keyword mentions across Xiaohongshu, Douyin, Weibo, and WeChat Official Accounts in real time.

Features Overview

Key: No official platform APIs required — pure Playwright scraping of public content.

Quick Start

Add Keyword Monitoring

User: Monitor keyword "brand_name" on Xiaohongshu and Douyin
User: Add sentiment monitoring for "product_name", platforms: Weibo + WeChat Official Account

→ Parse keyword and platforms → Create monitoring task → Execute first crawl → Return result summary

View Sentiment Report

User: Show sentiment report for "brand_name"
User: How is "competitor_name" trending in the last 7 days?

→ Return structured report: total mentions, positive/neutral/negative ratios, trending charts, top post list

Set Alert Rules

User: Set negative alert for "brand_name", threshold 10 posts/day, notify me when exceeded
User: Configure Feishu alert, push to "Operations Group"

→ Configure negative threshold and push channel → Auto-judge after each crawl

Manage Monitoring Tasks

User: List my sentiment monitoring tasks
User: Delete monitoring for "brand_name"
User: Pause monitoring for "competitor_name"

Tiered Features

Platform Monitoring Details

Xiaohongshu (XHS)

• Search URL: https://www.xiaohongshu.com/search_result?keyword={keyword}&source=web_explore_search
• Anti-detection: Playwright headless, UA rotation, random delay 3~8s
• Content extracted: Note title, body, author, likes/bookmarks/comments count, publish time

Douyin

• Search URL: https://www.douyin.com/search/{keyword}
• Anti-detection: Playwright headless, scroll simulation, lazy-load handling
• Content extracted: Video title, author, likes/comments/shares count, publish time

Weibo

• Search URL: https://s.weibo.com/weibo?q={keyword}&typeall=1
• Anti-detection: Playwright headless, UA rotation
• Content extracted: Post body, author, reposts/comments/likes count, publish time

WeChat Official Accounts

• Search URL: https://weixin.sogou.com/weixin?type=2&query={keyword}
• Anti-detection: Playwright headless
• Content extracted: Article title, abstract, account name, read count, publish time

Sentiment Analysis

Chinese semantic sentiment analysis via GLM-4 API:

Input: Post body / comment content
Output:
  sentiment: "positive" | "neutral" | "negative"
  score: -1.0 ~ 1.0 (negative to positive)
  reason: Brief reason summary

Classification rules:

• Positive: score > 0.1
• Neutral: -0.1 <= score <= 0.1
• Negative: score < -0.1

Alert Rules

Feishu Alert Message Template

Sentiment Alert | {keyword}
Time: {time}
Today's Negatives: {negative_count} (threshold: {threshold})
Negative Rate: {negative_rate}%
Latest Negative Posts:
- {title} — {platform} @{author}

Usage Examples

Example 1: Brand Sentiment Monitoring

User: Monitor "coffee brand" on Xiaohongshu and Douyin, crawl every day at 9am

→ Create task → Return confirmation → Next Cron trigger executes first crawl

Example 2: Competitor Negative Alert

User: Alert me via Feishu when negative posts appear for "competitor"

→ Set negative threshold alert → Configure Feishu group bot → Auto-push when threshold exceeded

Example 3: Sentiment Report

User: Generate this week's sentiment report for "brand_name"

→ Query local SQLite for this week's data → AI generate summary → Return Markdown report

Core Scripts

See scripts/sentiment.py for full implementation:

from scripts.sentiment import SentimentCompass

compass = SentimentCompass(tier="PRO")

# ─── Add keyword monitoring ──────────────
compass.add_keyword(
    keyword="brand_name",
    platforms=["xhs", "douyin", "weibo", "wechat"],
    frequency="daily",      # 6h/12h/daily/weekly
    priority=1,             # 1=high priority (Pro only)
)

# ─── Execute crawl (manual) ──────────────
results = compass.crawl_keyword("brand_name")

# ─── Sentiment analysis (single) ─────────
analysis = compass.analyze_sentiment("This product is really great, highly recommended!")
# → {"sentiment": "positive", "score": 0.85, "reason": "Contains positive words like 'great' and 'highly recommended'"}

# ─── Batch analysis (save API calls) ─────
batch = compass.batch_analyze([
    "Product is great, worth buying",
    "Quality is terrible, not worth the price at all",
    "It's okay, just average",
])
for item in batch:
    print(f"[{item['sentiment']}] {item['text'][:30]}")

# ─── Generate report ─────────────────────
report = compass.generate_report(keyword="brand_name", days=7)
print(report["summary"])   # AI-generated text summary
print(report["stats"])      # Statistical data

# ─── Check alerts ───────────────────────
alerts = compass.check_alerts(keyword="brand_name")
if alerts:
    compass.send_feishu_alert(alerts)

# ─── List tasks ─────────────────────────
tasks = compass.list_tasks()
for t in tasks:
    print(f"  {t['keyword']} — {t['platforms']} — {t['status']}")

Technical Implementation

• Crawler: Playwright (headless) for dynamic pages, UA rotation, random delay 3~8s
• AI Analysis: GLM-4 API (open.bigmodel.cn), batch analysis to save tokens
• Storage: SQLite (/tmp/sentiment-analysis-monitor/data.db) + JSON config
• Scheduling: OpenClaw Cron, supports 6h/12h/daily/weekly frequency
• Push: Feishu group bot Webhook / Email SMTP

Data Model

-- Monitoring tasks
CREATE TABLE tasks (
    id INTEGER PRIMARY KEY,
    keyword TEXT UNIQUE,
    platforms TEXT,           -- comma-separated: xhs,douyin,weibo,wechat
    frequency TEXT DEFAULT 'daily',
    priority INTEGER DEFAULT 0,
    status TEXT DEFAULT 'active',
    created_at TEXT,
    last_crawl_at TEXT
);

-- Post data
CREATE TABLE posts (
    id INTEGER PRIMARY KEY,
    keyword TEXT,
    platform TEXT,            -- xhs/douyin/weibo/wechat
    post_id TEXT,
    title TEXT,
    content TEXT,
    author TEXT,
    author_id TEXT,
    likes INTEGER DEFAULT 0,
    comments INTEGER DEFAULT 0,
    shares INTEGER DEFAULT 0,
    published_at TEXT,
    fetched_at TEXT,
    url TEXT UNIQUE
);

-- Sentiment analysis results
CREATE TABLE analyses (
    id INTEGER PRIMARY KEY,
    post_id INTEGER REFERENCES posts(id),
    sentiment TEXT,            -- positive/neutral/negative
    score REAL,                -- -1.0 ~ 1.0
    reason TEXT,
    analyzed_at TEXT
);

-- Alert records
CREATE TABLE alerts (
    id INTEGER PRIMARY KEY,
    keyword TEXT,
    alert_type TEXT,           -- threshold/trend
    threshold INTEGER,
    negative_count INTEGER,
    negative_rate REAL,
    triggered_at TEXT,
    notification_sent INTEGER DEFAULT 0
);

FAQ

Tier Limits

TIER_LIMITS = {
    "FREE":  {"max_keywords": 1,  "platforms": ["xhs"],                     "daily_limit": 50,  "history_days": 7},
    "PRO":   {"max_keywords": -1, "platforms": ["xhs","douyin","weibo","wechat"], "daily_limit": -1, "history_days": -1, "report": True, "priority": True},
}

Security Notes

• SSRF Protection: fetch_page() validates all URLs before sending to Playwright. Blocks: non-HTTP(S) schemes, localhost, 127.0.0.1, private IP ranges (10.x.x.x, 172.16-31.x.x, 192.168.x.x), link-local (169.254.x.x including AWS metadata 169.254.169.254), and IPv6 localhost. Unsafe URLs return None and log a warning.
• Subprocess execution: Uses node -e subprocess for Playwright browser automation (list form, not shell=True). curl calls for AI API use list form with json= payload. No command injection risk.
• Data storage: Uses /tmp/ for SQLite DB and config (no home directory write).
• Billing data: FEISHU_USER_ID transmitted to skillpay.me/api/v1/billing for per-call charging.

Billing

• Pay-per-call: $0.0100 USDT per execution via SkillPay.me
• Balance insufficient: Payment URL returned — user tops up at https://skillpay.me/sentiment-analysis-monitor
• External data flow: FEISHU_USER_ID transmitted to skillpay.me/api/v1/billing for balance charging

Required Environment Variables