skill-comment-insights · Comment Quantitative Analysis
Quantitative insight into a batch of comments: sentiment analysis (positive/neutral/negative share + representative comments), high-frequency word and phrase extraction, and intent mining for demands/complaints/questions — data for content reviews and topic feedback.
Built on jieba segmentation + SnowNLP sentiment + a social-media sentiment lexicon; rule-based and reproducible. Sentiment is approximate — read trends and shares; small samples (<20 comments) are flagged as limited. Analysis only, no replies.
Example invocation: "Analyze what this note's comment section is saying."
Full brief
Positioning
skill-comment-insights answers "what is the comment section saying": quantitative data analysis of comments. Analysis only — no replies, no crisis handling.
Core capabilities
- Sentiment analysis: positive/neutral/negative share + representative comments; SnowNLP baseline corrected with a social-media lexicon (internet slang like 绝绝子/yyds/避雷/翻车) for social-text bias.
- High-frequency words/phrases: jieba segmentation with POS filtering (nouns/verbs/adjectives kept) and stopword removal — informative words only.
- Intent mining: demands (asking for links/tutorials/same-item), complaints (product/service issue signals), questions — rule-matched, counted, with examples.
- Report output: JSON report (sentiment distribution/representative comments/top words/intent counts) + terminal-readable summary.
- Application guide: high negatives → locate the problem; top words → word clouds; demands → next topics; complaints → reviews.
Workflow
- Prepare comment data (txt/json/csv)
- Run
comment_insights.py analyzewith top-N and output path - Get the report JSON + terminal summary
- Apply per guide: locate issues, feed topics, feed reviews
Inputs & outputs
| Input | Notes |
|---|---|
| Comment data | .txt (one per line) / .json (array or object array) / .csv (specify column) |
Output: outputs/<topic>/report.json (sentiment/representative comments/top words/intent counts with examples) + terminal summary.
Boundaries with adjacent skills
| Skill | Lane |
|---|---|
| skill-comment-insights (this) | "What the comments say" (quantitative data) |
| skill-community-ops | "How to reply + crisis response" (ops actions) |
| skill-xhs-comment-reply | Comment scraping |
| skill-xhs-analyzer | Xiaohongshu viral patterns/keyword matrix (scraping lives there) |
Fit
- Comment-section reviews after notes/videos publish
- Mining user demands to feed back into topics
- Negative early-warning for product reputation
- Top words feeding word-cloud visuals
Before you start
- Needs
pip install jieba snownlp(jieba builds its dictionary cache on first run). - Sentiment is approximate: read trends and shares, not per-comment precision; human-check representative comments for key decisions; samples under 20 comments have limited share significance.
skill-comment-insights is part of the Aiglade Skill library. Invoke it from the Aiglade chat box in plain language.