SuperAPI AI Directory

community-analyzing-theme-comment

Data & analytics

View repository

Updated 10-16 04:31

About this skill

Skill: analyzing-theme-comment (deep comment analysis)

Overview

This skill performs deep analysis of comment threads for a specified post or video in the game community, following a structured flow to produce a report covering trends, viewpoint distribution, and synthesis.

Prerequisites

  • Tool: ae-community (ae-cli community commands)
  • gameId: Must be explicit
  • Channel: Know which channel hosts the target post/video

When to use

  • Sentiment in a hot post’s comments
  • Audience feedback after a video drop
  • Viewpoint distribution for a specific topic
  • Evaluating content performance and player mood

Execution flow

Step 1: Identify the target post or video

1.1 Check whether the user specified a target

Confirm the user provided:

  • Post/video UUID
  • channelId
  • Resource type: post (0) or video (1)

If yes, go to Step 2.

1.2 Search when no target is given

Use ae-cli community +search_posts:

ae-cli community +search_posts \
  --space-id <space_id> --game-id <game_id> \
  --start-time <yyyy-MM-dd> --end-time <yyyy-MM-dd> \
  --search-word "<keyword>" \
  --channel-id-list 1,2 \
  --resource-type 0,1 \
  --order-by 4 --page-num 1 --page-size 100

Key parameters:

Flag Required Notes
--space-id Yes Space ID
--game-id Yes Game ID
--start-time Yes Lower bound, yyyy-MM-dd
--end-time Yes Upper bound, yyyy-MM-dd
--search-word No Keyword
--channel-id-list No Comma-separated channel IDs, for example 1,2
--resource-type No Comma-separated types: 0 post, 1 video; for example 0,1
--order-by No Use 4 to sort by heat descending
--page-num No Page number, starting at 1
--page-size No Page size, maximum 10000

Channel reference:

channelId Name
1 TapTap
2 BiliBili
3 Douyin
5 Weibo
7 RedNote
8 HeyBox
9 Douyu
11 Tieba
17 Nga

Logic:

  1. Search by keyword or hot topic
  2. Sort by heat; surface top items
  3. Have the user pick a target
  4. Capture UUID and channelId

Step 2: Fetch comments

2.1 Post/video detail

Use ae-cli community +get_post_detail:

ae-cli community +get_post_detail \
  --space-id <space_id> --game-id <game_id> \
  --channel-id <channel_id> --uuid <content_uuid> \
  --resource-type <0_or_1> \
  --reply-order-by 0 --reply-page-num 1 --reply-page-size 100
Flag Required Notes
--space-id Yes Space ID
--game-id Yes Game ID
--channel-id Yes Channel ID
--uuid Yes Content UUID
--resource-type Yes 0 post, 1 video
--reply-order-by No 0 newest first, 1 oldest first
--reply-page-num No Comment page number, starting at 1
--reply-page-size No Comment page size, maximum 10000

2.2 Paginate to completeness

  1. Fetch the first page with --reply-page-num 1 --reply-page-size 100.
  2. If the response indicates more comments, rerun the same command with --reply-page-num incremented by one.
  3. Repeat until the response indicates no next page; do not claim completeness before then.
  4. For video danmu, paginate independently with --danmu-page-num and --danmu-page-size when danmu is part of the requested analysis.

Filter out low-value replies:

  • Spam / duplicates / flood
  • Emoji-only or symbol-only (“???”, “!!!”)
  • Fewer than 3 meaningful characters
  • System/automated official messages

Record:

  • Total vs. valid comment counts
  • Time span
  • Channel distribution if applicable

Step 3: Deep analysis

3.1 Trend

3.1.1 Time distribution (daily)

Date Comments Cumulative
2026-03-10 150 150
… … …

3.1.2 Sentiment by day

Date Positive Neutral Negative
2026-03-10 0.45 0.35 0.20
… … … …

3.2 Viewpoints

3.2.1 Method

  1. Cluster themes from valid text
  2. Surface recurring keywords/phrases
  3. Merge near-duplicates
  4. Split positive vs. negative stances

3.2.2 Viewpoint object

Field Description
Title Short label (~20 chars)
Description ~100 chars
Match count Comments aligned with this view
Examples Up to 10 representative quotes

3.2.3 Opposing views

  • Allow clear pro/con on the same subject
  • Do not merge contradictions; list separately

Example:

Positive viewpoint 1

  • Title: Roguelike mode praised
  • Description: Players like the new roguelike loop for variety and replay value.
  • Matches: 45
  • Quotes: …

Negative viewpoint 1

  • Title: Daily grind frustration
  • Description: Some players say dailies take too long and mats are too scarce for low spenders.
  • Matches: 38
  • Quotes: …

3.3 Synthesis

3.3.1 Volume summary

  • Total vs. valid counts
  • Overall sentiment mix
  • Peaks and likely drivers
  • Optional benchmark vs. similar content

3.3.2 Viewpoint synthesis

  • Dominant positive themes and reach
  • Dominant negative themes and reach
  • Conflict axes
  • Core asks from players

3.3.3 Ecosystem

  • Tone: harmonious / polarized / neutral
  • Alignment with OP/video
  • Audience segments and risks/opportunities

Output format

# Themed Comment Deep-Dive Report

## 1. Basics
- Title: [title]
- Channel: [name]
- Published: [time]
- Analysis window: [start] – [end]
- Valid comments: [X]

## 2. Trends
### 2.1 Time series
[table or narrative]
### 2.2 Sentiment mix
[table or narrative]

## 3. Viewpoints
### 3.1 Positive
[…]
### 3.2 Negative
[…]

## 4. Summary
### 4.1 Volume
### 4.2 Viewpoints
### 4.3 Ecosystem & recommendations

Notes

  1. Completeness: Paginate until all comments are retrieved.
  2. Quality: Strictly filter spam/noise.
  3. Objectivity: Ground clusters in actual text.
  4. Opposition: Preserve real splits in opinion.
  5. Representative quotes: Weight typicality, engagement, recency.

Error handling

Situation Action
Invalid gameId Ask user to verify
Missing post/video Verify UUID
Fetch failure Retry; check connectivity
Fewer than 10 valid comments Warn sample is thin; suggest another target
Invalid channelId Show channel list

Tool dependencies

Tool Role
ae-cli community +search_posts Find posts/videos
ae-cli community +get_post_detail Detail + comments
ae-cli community +get_comments_summary Optional assist