xhs-trend-scan

name

xhs-trend-scan

description

Detects emerging concept words on Xiaohongshu (小红书) before they go mainstream — extracts 2-4 char Chinese n-grams from recent viral notes, filters against a 5000-word common lexicon baseline, and ranks by recency × viral note appearance × time concentration. Surfaces words like 「奥德赛时期」 1-2 weeks before mainstream pickup. Use when the user asks to 找新概念词 / 流行新词 / 哪些词刚开始火 / 新冒出来的词 / 早期热词识别 / trend scan / emerging concepts.

xhs-trend-scan — 流行新概念词识别

Mines TOP 60-90 recent viral notes for emerging vocabulary that hasn't yet gone mainstream — the 「2 weeks before everyone uses it」 advantage.

When to use

User asks:

• 「本周有哪些新冒出来的概念词？」
• 「找出近 14 天才开始火的词」
• 「哪些词我应该早点用？」
• 「Find emerging concepts on xhs this week」

Quick start

python3 shared/lib/cli.py trend-scan \
  --keywords 情绪管理,内耗,自我疗愈 \
  --window-days 14

If --keywords omitted, uses config.yaml::defaults.keywords. If you've recently run xhs-viral-pulse, this skill auto-reuses that data (saves API calls).

What it produces

Single docx in Feishu folder with 4 sections:

1. TOP 10 emerging concepts — table with 词 / first appearance date / freq / appearing in viral notes / score
2. For each concept: source note examples (showing how the word is used in context)
3. Trend trajectory — likes growth curve per concept
4. Action推荐 — 3 concepts most worth Joyce adopting in next-week content

Detection algorithm (key design)

score = (
    log(frequency_in_top_viral) * 3        # appears in viral notes = strong signal
  + recency_concentration * 2              # most appearances within last N days
  + cross_keyword_appearance * 2           # appears across multiple keyword searches
  - in_common_lexicon * 10                 # heavy penalty if it's a common Chinese word
  - in_existing_brand_lexicon * 5          # penalty if it's a brand/proper noun
)

Ranked descending; take TOP 10 with score > threshold.

Common lexicon

5000-word Chinese baseline at shared/data/common_lexicon.txt — ensures we surface NEW words, not just common ones. Built from public Chinese frequency tables.

Workflow detail

See workflow.md.

Composition with other skills

• Pair with xhs-viral-pulse to immediately incorporate trend words into the content plan
• Pair with xhs-viral-rewrite to use trend words as voice signature additions

Failure modes — feishu output is REQUIRED

The skill must end with a Feishu doc URL handed back to the user. If feishu push fails for any reason, stop and ask the user to fix the configuration. Do NOT silently degrade to local-only output.

Failure	Action
lark-cli binary not in PATH	STOP. Tell user to install lark-cli per setup/02_install_lark_cli.md.
lark-cli auth expired (LarkAuthExpired)	STOP. Tell user: lark-cli auth login --as bot
Bot lacks scope (LarkScopeMissing)	STOP. Show the scope_url from the error; user grants in 1 click in 飞书 console.
config.yaml missing feishu.folder_token	STOP. Tell user to edit config.yaml.

Local markdown / JSON files saved during the run are intermediate artifacts — they are not the deliverable.

Examples

See examples.md and ../../examples/trend_scan_sample.md.