SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsshengjiang-social-media-research

shengjiang-social-media-research

Writing & research

View repository

Updated 10-15 19:58

About this skill

Ginger Social Media Research

Turn "search a few pieces of content and take a look" into social media research with sources, raw data, and structured results.

First, clarify the capability boundaries

This Skill does not include social media data sources, nor does it provide or resell APIs.

  • The automated route calls the third-party TikHub API. Users need to register, purchase, or use its trial quota themselves, and configure their own TIKHUB_API_KEY. Fees, quotas, supported interfaces, and platforms are subject to TikHub's current official documentation.
  • The free route uses the free features of the third-party "Social Media Assistant" browser extension. Users need to install the extension themselves, log in to the platform, click to collect, and export Excel; the AI is responsible for teaching the steps, reading files, cleaning, deduplicating, analyzing, and generating reports.
  • Do not bypass login, payment, CAPTCHAs, access controls, or platform restrictions. Only collect data that users have the right to access and that complies with platform rules.

When users ask "is it free," "whose API is it," "where does the Key come from," or "can I do it without an API," directly state the above facts, without ambiguity, and do not describe third-party capabilities as Ginger's self-built interfaces.

Choose a route

First clarify the goal, platform, target, time range, intended use of results, and expected fields, then choose the lowest-cost feasible route:

  1. The user already has a TikHub Key and clearly needs batch or automation: take the paid API route.
  2. The user wants it free, has no Key, is only researching a small number of targets, or is willing to operate manually: take the Social Media Assistant route.
  3. The user has already provided Excel, CSV, JSON, a link list, or screenshots: go directly to organizing and analysis, and do not require re-collection.
  4. The platform or fields are not supported by the current tools: clearly mark "to be collected later," provide alternatives, and do not fabricate data.

By default, give an execution preview first, and do not directly generate large batch requests that may incur charges.

Standard workflow

1. Define the research question

Converge vague needs into verifiable questions, for example:

  • Find 20 AI hands-on accounts worth learning from;
  • Break down the 50 works with the highest engagement in the past 30 days;
  • Extract beginners' real questions from 1,000 comments;
  • Collect titles, openings, structures, and user feedback evidence for a certain topic.

At minimum, record:

  • Platform and target links;
  • Time range and sample size;
  • Required fields;
  • Final deliverables;
  • Whether third-party paid calls are acceptable.

2. Provide a collection preview

The preview must include:

  • The chosen route and the reason;
  • Planned requests or manual operations;
  • Sample size and possible cost;
  • Output directory and fields;
  • Known gaps and risks.

For the paid route, first do 1–3 samples. After the sample fields are correct, then let the user confirm the batch scope. Do not promise a fixed price.

3. Collect and preserve evidence

It is recommended to save according to the following directory:

social-research/
├── raw/            # 原始响应或原始导出，不覆盖
├── normalized/     # 统一字段后的 CSV / JSON / Excel
├── transcripts/    # 平台字幕或第三方 ASR 逐字稿
├── evidence/       # 截图、链接清单与引用证据
└── reports/        # 分析报告、选题表和对标卡

Each item must retain at least platform、source_url、author_name、published_at、collected_at. For field definitions, see references/output-schema.md。

4. Cleaning and Structuring

  • Deduplicate using the platform content ID or a normalized URL;
  • Keep raw fields separate from derived fields;
  • Keep missing values empty or note the reason; do not fill in guessed values;
  • For numbers such as plays, likes, comments, and saves, retain the collection time;
  • Keep comments and usernames only to the minimum scope needed to complete the task;
  • Clearly separate external content, user judgments, and AI conclusions.

5. Obtain the Transcript

The priority order is fixed as:

  1. The platform's official subtitles or text provided by the author;
  2. A transcript the user already has;
  3. A third-party ASR API configured by the user.

When no usable source exists, mark it as "transcript pending" and do not use local Whisper or other local speech models to pass off recognition as completed.

6. Analysis and Delivery

Conclusions must be traceable back to the original link or file. Recommended deliverables:

  • Account sample table;
  • Content and data detail table;
  • Clustering of comment questions and needs;
  • Breakdown of titles, hooks, structure, and presentation methods;
  • Actionable topic list;
  • Data limitations and items pending collection.

Do not give only vague conclusions like "the content is good" or "engagement is great."

Paid API Route

Before execution, read in full references/paid-api-route.md。

Call principles:

  • Have the user set the Key in a local environment variable; do not let the user send the Key into the chat;
  • Use only endpoints confirmed to exist in the official TikHub documentation;
  • Use scripts/tikhub_request.py --dry-run to check the request first;
  • Collect 1–3 items first, then batch;
  • Do not write the Key into logs, reports, command output, or Git;
  • When receiving 401, 402, 429, or platform restrictions, state it truthfully and give the next step.

Free Manual Route

Before execution, read in full references/free-manual-route.md。

The AI's role is to guide the user step by step through:

  1. Installing and pinning the "Social Media Assistant" extension from the official page;
  2. Opening and logging in to the target platform;
  3. Entering the search, account, content, or comment page;
  4. Selecting the collection scope and fields in the extension, starting with a small sample;
  5. Exporting to Excel;
  6. Giving the Excel file to the AI for cleaning and analysis.

It must be made clear: installation, login, and collection clicks in the browser must be completed manually by the user. The features and limitations of the free version may change; the official page shall prevail.

Security and Compliance

  • Do not collect passwords, Cookies, session tokens, payment information, or unrelated personal information;
  • Do not let users paste API Keys; recommend using environment variables of the current terminal session;
  • Do not upload .env, configurations, responses, or screenshots containing real keys;
  • Do not publicly disclose signature parameters in temporary download addresses;
  • Do not perform collection involving harassment, profiling discrimination, privacy infringement, or bypassing platform mechanisms;
  • When citing third-party content, retain the source and comply with copyright and platform rules;
  • Before publishing externally, check whether usernames, comment text, and personal information are truly necessary.

Failure Handling

  • No API Key: switch to the free route; do not treat it as a blocker;
  • The free extension does not support the current platform or fields: have the user export platform backend data, provide a link list, or choose an officially permitted third-party service;
  • The data volume is too large: batch it by platform, time, or content type;
  • Field drift: keep the original file, update the mapping, and do not modify the original data;
  • Subtitles cannot be obtained: deliver metadata and content analysis, and list the verbatim transcript as pending.