---
document_id: agent.cookbook.social_content_analysis
schema_version: 2
parent_document_id: agent.cookbook
section: cookbook
---

<!-- Generated by scripts.build_agent_guide; do not edit. -->

# Social content queries and analysis / 社媒内容查询与分析

Find relevant information on social platforms, including original posts, comments, and follow-up discussions. Choose platform-specific search conditions and collection scope from the request; for bulk work, validate a small pilot before collecting the agreed scope. / 从社交平台查找相关信息，获取原始内容、评论和后续讨论。按问题选择平台、检索条件和采集范围；批量任务先试抓验证，再按约定范围采集。

## Business questions / 商业问题

- In which situations do people use this type of product, including unexpected uses?
  用户在什么场景下使用这类产品，有哪些意料之外的用途？
- What situations cause neck-pillow discomfort, such as neck squeezing, and how do users work around it?
  用户吐槽 U 型枕“卡脖子”时，具体在什么场景，后来怎样解决？
- Why do people choose or reject this product, and what alternatives do they use?
  用户为什么选择或放弃这个产品，有哪些购买顾虑和替代方案？
- What are people discussing about this brand, product, or industry recently, and which posts are driving the public concerns or controversy?
  围绕这个品牌、产品或行业，最近大家在讨论什么？舆情焦点和争议来自哪些内容？
- How do concerns about the same event differ across platforms, and how is negative discussion spreading?
  同一事件在不同平台的关注点有何不同，负面讨论怎样传播？
- Track new posts and comments for these terms or accounts and flag important changes, with a recurring execution arrangement.
  按这些关键词或账号跟踪新增内容和评论，出现重要变化时提醒我。
- Collect the agreed period of product-use posts and comments after a pilot, to investigate unmet needs.
  先试抓，再按约定时期采集使用体验和评论，挖掘未满足需求。

## Approach / 方法

1. Choose the actions the task needs; these are not mandatory sequential stages. Reuse the supplied subject, platforms, period, and content requirements. Exact references can be queried directly. For bulk work, retain a clear scope; otherwise propose a workable scope and clarify only conditions that change execution. Do not require a full questionnaire or repeat settled confirmations.
   按任务选用做法，不要求逐项执行。沿用已有对象、平台、时期和内容要求，精确引用可直接查。批量任务范围明确就沿用；不明确时先给出可行范围，只补问影响执行的条件，不要求填完整问卷或重复确认。
2. Platforms and supported conditions. Domestic examples include Xiaohongshu, Douyin, Weibo, Bilibili, Zhihu, Kuaishou, WeChat Official Accounts, and WeChat Channels; overseas examples include X (Twitter), TikTok, YouTube, Instagram, Facebook, Reddit, and LinkedIn. Choose platforms as needed, not all of them or equal quotas. Capabilities come from authenticated discover; read new operation contracts and reuse applicable known contracts. Check inputs, allowed values, defaults, and pagination; do not copy names, values, cursors, or syntax across platforms. Divide work by actual accounts, language, and formats, retaining the requested scope.
   适用平台与查询条件。国内包括小红书、抖音、微博、B站、知乎、快手、微信公众号、视频号；海外包括 X（Twitter）、TikTok、YouTube、Instagram、Facebook、Reddit、LinkedIn 等。平台按需求选，不要求全查或平均分配。各平台能力不同，具体以认证 discover 为准；新操作先读契约，已知且适用的契约直接复用。核对入参、可选值、默认值和分页，不跨平台照搬参数名、取值、游标或搜索语法；按实际账号分布、语言和内容形态分工，保留用户要求的范围。
3. Build useful search terms rather than a fixed checklist. Select relevant clues about people (including KOLs), objects, events, relationships, time, or place; try names, handles, aliases, hashtags, neutral descriptions, and everyday wording as needed. Broad topics may need people, use-case, or problem terms. Do not search only accusations or a presumed conclusion; add responses, clarifications, or follow-ups when relevant. Use phrase or exclusion syntax only where supported.
   按问题拟词，不设固定词表。按需选择人物（如 KOL）、对象、事件、关系、时间地点线索，以及本名、账号名、别称、标签、中性描述和日常说法；不要求每次覆盖所有词类。宽泛话题可加人群、场景或问题词，不只搜指控或预设结论，必要时再查回应、澄清和后续；短语、排除词等语法仅在平台支持时用。
4. Pilot and rewrite per platform when the query or conditions need validation. Reuse a still-valid pilot of the same scope rather than repeating it. Check relevance, dates, filters, required fields, and pagination with a few terms or pages; the pilot is not the final collection limit. For pain-point research, inspect the density of concrete experiences; for public reactions, inspect relevant sources and viewpoints. Do not choose solely by engagement or comment counts, or universally discard ads, reposts, or link requests. Use this check to refine terms or choose analysis samples, not silently reduce an agreed collection scope. Narrow broad results, try aliases for sparse results, or trace originals when results are mostly reposts; for latest updates consider time sorting, and for attention consider engagement, where supported.
   试搜后逐平台改写。查询或条件需要验证时少量试抓，已有同范围且仍有效的试抓可复用。检查相关性、日期、筛选、字段与分页；试抓量不是正式采集上限。痛点研究看具体经历的密度，舆情研究看相关来源和观点，不按热度或评论总数机械选取，也不一概删除广告、转述或求链接。质量检查用于调词或选分析样本，不擅自缩小约定采集范围。太泛就加线索，太少就换别称，多为转载可追原话；查最新动态可优先按时间，研究关注度可看高互动，按平台能力选择。
5. Collect the required layers after validation. Read bodies for statements and context, comments for reactions, and replies for the exchange. Inspect returned fields first and reuse content already present. For targeted or exploratory work, deepen relevant posts as needed. For bulk work, continue searches and pagination for each agreed term, account, and platform, and retrieve the agreed bodies, comments, or replies instead of retaining only a few examples. Verify account identity when needed, but do not treat it as proof of a claim or reconstruct missing content from snippets.
   按需补正文、评论和回复。原话与上下文读正文，反应取评论，对话取回复，已返回的内容直接复用。定向查询或探索按需深入相关内容；批量采集在试抓验证后，按约定的检索词、账号和平台逐项搜索、分页，并补齐约定的正文、评论或回复，不能只保留几个代表样本。必要时核对账号身份，但身份本身不能证实消息，也不能用摘要补写缺失内容。
6. Trace sources when the task requires checking news or disputed facts; a collection-only task does not require adjudicating every item first. Follow citations to original posts or speakers. Check event time, publication time, and later updates, and compare statements, responses, and independent evidence across relevant platforms. Reposts of the same source are not independent corroboration. Distinguish firsthand claims, retellings, speculation, and verified facts; being early or firsthand is not proof.
   需要核实消息或争议事实时再溯源，单纯采集不以逐条鉴定真伪为前提。沿引用找原帖、当事人或原始出处，核对事件时间、发布时间和后续更新，结合相关平台的原话、回应及独立证据；同源转载不算独立佐证。区分一手说法、转述、猜测和已核实内容，发得早或自称亲历不等于属实。
7. Advance and resume within the task scope. Exploratory work may stop when the question is answered. Bulk work continues through the agreed reachable pages or until user, budget, or service limits; repeated viewpoints do not justify early completion. For bulk or resumable tasks, save results, completed IDs, and continuation positions in batches using existing tool results or local files. Reuse successes, fill only confirmed unfinished work, and read the original task when an outcome is unknown. Record gaps and limits. Deduplicate by platform plus content or comment ID and retain shared provenance; compare platforms with their conditions, periods, sorting, and samples visible. Follow current approval, pagination, and recovery contracts.
   按范围推进与续取。探索可在问题已回答时收敛；批量按约定范围取完可达的分页，或到达用户、预算、服务上限，不能因观点重复就提前停。批量或需继续的任务分批保存结果、已完成 ID 和续取位置，复用工具结果或本地文件；中断后只补确认未完成项，结果未知时先读原任务。记录缺口和限制。按平台加内容或评论 ID 去重、保留共同来源；跨平台比较保留条件、时期、排序和样本差异。审批、分页和恢复按契约。

## Boundaries / 边界

- This method supports one-time inquiries and the collection steps within public-opinion monitoring, across currently available domestic and overseas platforms; it is not limited to consumer goods. E-commerce buyer reviews follow their own guide. Monitoring additionally needs a defined frequency, changes of interest, historical results, and an actual execution and notification arrangement; a one-time search does not start automatic monitoring.
  本方法用于当前可用国内、海外平台的一次性调查及舆情监控中的取数，不限消费品；电商买家评价另走相应指南。监控还需明确频率、关注变化和提醒条件，并有历史结果、持续执行与通知安排，不能把一次检索称为已开启自动监控。
- Complete collection means reachable results under the chosen conditions and current interface, not all platform content. Samples do not represent all public opinion; engagement proves neither truth, firsthand use, purchase intent, nor sales. When counts or coverage are needed, distinguish search hits, bodies obtained, comments/replies, and records used in analysis. Keep raw results and do not call search-hit counts valid-feedback counts. Distinguish no matches, missing content, and failures; do not call a limited or interrupted run complete.
  “抓全”只指所选条件下当前接口可获取的结果，不保证整个平台无遗漏。样本不代表全量舆情，热度不等于真实使用、购买意愿或销量。需要统计或说明覆盖时，分别记录搜索条数、取得正文数、评论/回复数和分析采用量，保留原始结果，不把搜索条数当成有效反馈数。区分无匹配、内容缺失与失败，不把受限或中断的采集称为已抓全。
- When evidence is insufficient, keep a claim unverified or disputed; do not make gossip factual because it repeats across platforms. Record actual publication dates and coverage limitations; collection time is not publication time. Use only fields and capabilities confirmed by the runtime.
  证据不足就保留未证实或有争议状态，不能因多个平台都在传就把八卦当事实。记录实际发布时间和覆盖限制，采集时间不能代替发布时间；只使用运行时确认的能力和字段。
