Back to Details

analyzing-twitter-sentiment-for-topic

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Analyzing Twitter Sentiment for a Topic

分析某话题的Twitter情绪

Collects a sample of tweets about any topic or keyword and performs sentiment analysis across the dataset. Identifies dominant emotional tone, key themes driving positive/negative sentiment, and volume patterns over time.
收集任意话题或关键词的推文样本，并对数据集进行情绪分析。识别主导情感基调、驱动正面/负面情绪的关键主题，以及推文数量随时间变化的模式。

Prerequisites

前提条件

  • APIFY_TOKEN
    environment variable set
  • Optional: Apify MCP server installed
  • 已设置
    APIFY_TOKEN
    环境变量
  • 可选：已安装Apify MCP服务器

Inputs

输入参数

ParameterTypeRequiredDefaultNotes
searchTerms
array✅
[]
Twitter advanced search queries (e.g. 
["#AI lang:en", "from:NASA"]
)
sort
stringOptional
Top
Sort order: 
Latest
, 
Top
, or 
Latest+Top
tweetLanguage
stringOptional—ISO 639-1 language code (e.g. 
en
)
maxItems
numberOptionalUnlimitedMaximum tweets to return
onlyVerifiedUsers
booleanOptional
false
Only tweets from verified users
onlyTwitterBlue
booleanOptional
false
Only Twitter Blue subscribers
onlyImage
booleanOptional
false
Only tweets with images
onlyVideo
booleanOptional
false
Only tweets with videos
onlyQuote
booleanOptional
false
Only quote tweets
author
stringOptional—Filter to a specific author handle
inReplyTo
stringOptional—Tweets replying to a specific handle
mentioning
stringOptional—Tweets mentioning a specific handle
geotaggedNear
stringOptional—Tweets near a location
withinRadius
stringOptional—Radius around geotaggedNear
geocode
stringOptional—Lat/lng + radius string
placeObjectId
stringOptional—Tweets tagged with a place
minimumRetweets
numberOptional—Minimum retweet count
minimumFavorites
numberOptional—Minimum like count
minimumReplies
numberOptional—Minimum reply count
start
stringOptional—Tweets after this date (YYYY-MM-DD)
end
stringOptional—Tweets before this date (YYYY-MM-DD)
includeSearchTerms
booleanOptional
false
Add the matched search term to each tweet
customMapFunction
stringOptional—JavaScript function to transform each output object
参数类型是否必填默认值说明
searchTerms
数组✅
[]
Twitter高级搜索查询（例如：
["#AI lang:en", "from:NASA"]
）
sort
字符串可选
Top
排序方式：
Latest
（最新）、
Top
（热门）或
Latest+Top
（最新+热门）
tweetLanguage
字符串可选—ISO 639-1语言代码（例如：
en
）
maxItems
数字可选无限制返回的最大推文数量
onlyVerifiedUsers
布尔值可选
false
仅返回认证用户的推文
onlyTwitterBlue
布尔值可选
false
仅返回Twitter Blue订阅用户的推文
onlyImage
布尔值可选
false
仅返回带图片的推文
onlyVideo
布尔值可选
false
仅返回带视频的推文
onlyQuote
布尔值可选
false
仅返回引用推文
author
字符串可选—筛选特定作者的推文（需提供账号 handle）
inReplyTo
字符串可选—回复特定账号的推文
mentioning
字符串可选—提及特定账号的推文
geotaggedNear
字符串可选—特定地点附近的推文
withinRadius
字符串可选—
geotaggedNear
参数指定地点的半径范围
geocode
字符串可选—纬度/经度 + 半径的字符串
placeObjectId
字符串可选—标记了特定地点的推文
minimumRetweets
数字可选—最小转发量
minimumFavorites
数字可选—最小点赞量
minimumReplies
数字可选—最小回复量
start
字符串可选—此日期之后的推文（格式：YYYY-MM-DD）
end
字符串可选—此日期之前的推文（格式：YYYY-MM-DD）
includeSearchTerms
布尔值可选
false
为每条推文添加匹配的搜索词
customMapFunction
字符串可选—用于转换每个输出对象的JavaScript函数

Workflow

工作流程

Progress:
- [ ] Step 1: Define topic and sentiment scope
- [ ] Step 2: Collect tweets via search
- [ ] Step 3: Classify sentiment per tweet
- [ ] Step 4: Identify themes per sentiment bucket
- [ ] Step 5: Deliver sentiment report
进度：
- [ ] 步骤1：定义话题和情绪分析范围
- [ ] 步骤2：通过搜索收集推文
- [ ] 步骤3：对每条推文进行情绪分类
- [ ] 步骤4：按情绪类别识别主题
- [ ] 步骤5：交付情绪分析报告

Step 1: Clarify Parameters

步骤1：明确参数

Ask the user for:
  • Topic, keyword, or brand to analyze
  • Date range (default: last 7 days — Twitter sentiment data decays fast)
  • Language (default: English)
  • Sample size (default: 500 tweets — sufficient for reliable distribution)
  • Exclude retweets? (default: yes — reduces duplicated opinion signals)
  • Comparison topic (optional — for side-by-side sentiment comparison)
向用户确认以下信息：
  • 要分析的话题、关键词或品牌
  • 日期范围（默认：过去7天——Twitter情绪数据时效性强）
  • 语言（默认：英语）
  • 样本量（默认：500条推文——足以保证结果的可靠性）
  • 是否排除转发？（默认：是——减少重复的意见信号）
  • 对比话题（可选——用于情绪的对比分析）

Step 2: Collect Tweets

步骤2：收集推文

Recommended — run_actor.js (handles waiting, output, and file saving automatically):
bash
undefined
推荐方式 — 使用run_actor.js（自动处理等待、输出和文件保存）：
bash
undefined

Quick answer (prints table to chat)

快速输出（在聊天中打印表格）

node scripts/run_actor.js
--actor "apidojo~tweet-scraper"
--input '{"param": "value"}'
node scripts/run_actor.js
--actor "apidojo~tweet-scraper"
--input '{"param": "value"}'

Save as CSV

保存为CSV文件

node scripts/run_actor.js
--actor "apidojo~tweet-scraper"
--input '{"param": "value"}'
--output YYYY-MM-DD_results.csv --format csv
node scripts/run_actor.js
--actor "apidojo~tweet-scraper"
--input '{"param": "value"}'
--output YYYY-MM-DD_results.csv --format csv

Save as JSON

保存为JSON文件

node scripts/run_actor.js
--actor "apidojo~tweet-scraper"
--input '{"param": "value"}'
--output YYYY-MM-DD_results.json --format json
> `APIFY_TOKEN` must be set in environment or `.env` file.

**If Apify MCP is available:**
Tool: apify:run-actor Actor: "apidojo~tweet-scraper" Input: { "searchTerms": ["[TOPIC_KEYWORD]"], "maxItems": 500, "tweetLanguage": "en", "since": "[YYYY-MM-DD]", "until": "[YYYY-MM-DD]" }

**REST API fallback:**
```bash
curl -X POST \
  "https://api.apify.com/v2/acts/apidojo~tweet-scraper/runs?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "searchTerms": ["[TOPIC_KEYWORD]"],
    "maxItems": 500,
    "tweetLanguage": "en",
    "since": "[YYYY-MM-DD]"
  }'
node scripts/run_actor.js
--actor "apidojo~tweet-scraper"
--input '{"param": "value"}'
--output YYYY-MM-DD_results.json --format json
> 必须在环境变量或`.env`文件中设置`APIFY_TOKEN`。

**如果Apify MCP可用：**
工具: apify:run-actor Actor: "apidojo~tweet-scraper" 输入: { "searchTerms": ["[TOPIC_KEYWORD]"], "maxItems": 500, "tweetLanguage": "en", "since": "[YYYY-MM-DD]", "until": "[YYYY-MM-DD]" }

**REST API备选方案：**
```bash
curl -X POST \
  "https://api.apify.com/v2/acts/apidojo~tweet-scraper/runs?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "searchTerms": ["[TOPIC_KEYWORD]"],
    "maxItems": 500,
    "tweetLanguage": "en",
    "since": "[YYYY-MM-DD]"
  }'

Step 3: Classify Sentiment

步骤3：情绪分类

For each tweet's 
text
, classify as Positive, Negative, or Neutral using lexical signals:
Positive indicators: love, great, amazing, perfect, best, win, excited, congrats, excellent, recommend, beautiful, proud, happy, thank, awesome, incredible Negative indicators: hate, awful, worst, terrible, broken, scam, disappointed, angry, frustrated, disgusted, avoid, never again, shame, sad, fail, wrong, bad Strong negative amplifiers: "can't believe", "what a joke", "are you serious", "wtf", "this is ridiculous" Neutral default: Everything else
For ambiguous cases, use emoji signals:
  • 😍🥰❤️🙌👏✨🔥 → lean Positive
  • 😡🤬😤💀🗑️🤢👎 → lean Negative
  • 🤔😐🤷 → lean Neutral
Weight tweets by engagement: a tweet with 1,000 likes carries more signal than one with 0.
针对每条推文的
text
内容，通过词汇信号将其分类为正面、负面或中性：
正面指标： love、great、amazing、perfect、best、win、excited、congrats、excellent、recommend、beautiful、proud、happy、thank、awesome、incredible 负面指标： hate、awful、worst、terrible、broken、scam、disappointed、angry、frustrated、disgusted、avoid、never again、shame、sad、fail、wrong、bad 强负面强化词： "can't believe"、"what a joke"、"are you serious"、"wtf"、"this is ridiculous" 默认中性： 其他所有内容
对于模糊的情况，使用表情符号信号：
  • 😍🥰❤️🙌👏✨🔥 → 倾向于正面
  • 😡🤬😤💀🗑️🤢👎 → 倾向于负面
  • 🤔😐🤷 → 倾向于中性
按互动量为推文加权：一条获得1000个点赞的推文比0点赞的推文更具信号价值。

Step 4: Theme Extraction

步骤4：主题提取

For Negative tweets: identify the top 3-5 recurring nouns/themes. What are people upset about specifically? For Positive tweets: identify the top 3-5 recurring praise themes.
Look for proper nouns (people, places, products), specific events, or feature names that appear repeatedly.
对于负面推文：识别出现次数最多的3-5个名词/主题。人们具体在不满什么？ 对于正面推文：识别出现次数最多的3-5个获赞主题。
寻找重复出现的专有名词（人物、地点、产品）、特定事件或功能名称。

Step 5: Format Report

步骤5：格式化报告

Output Format

输出格式

undefined
undefined

Twitter Sentiment Analysis: "[TOPIC]"

Twitter情绪分析："[TOPIC]"

Period: [DATE_RANGE] | Tweets analyzed: [N] | Date: [DATE]
时间段：[DATE_RANGE] | 分析推文数量：[N] | 日期：[DATE]

Overall Sentiment

整体情绪

████████████░░░░░░░░  Positive: [X%] ([N] tweets)
████░░░░░░░░░░░░░░░░  Negative: [X%] ([N] tweets)
██████████░░░░░░░░░░  Neutral:  [X%] ([N] tweets)
Weighted by engagement:
  • Positive sentiment accounts for [X%] of total likes/RTs
  • Negative sentiment accounts for [X%] of total likes/RTs
Overall verdict: [Mostly Positive / Mixed / Mostly Negative / Polarized]
████████████░░░░░░░░  正面：[X%]（[N]条推文）
████░░░░░░░░░░░░░░░░  负面：[X%]（[N]条推文）
██████████░░░░░░░░░░  中性：[X%]（[N]条推文）
按互动量加权：
  • 正面情绪占总点赞/转发量的[X%]
  • 负面情绪占总点赞/转发量的[X%]
整体结论： [以正面为主 / 混合 / 以负面为主 / 两极分化]

Top Negative Themes

热门负面主题

  1. "[Theme]" — [N] tweets, [N] total likes Example: "@[handle]: [tweet excerpt]"
  2. "[Theme]" — [N] tweets
  3. "[Theme]" — [N] tweets
  1. "[主题]" — [N]条推文，总点赞量[N] 示例："@[handle]: [推文节选]"
  2. "[主题]" — [N]条推文
  3. "[主题]" — [N]条推文

Top Positive Themes

热门正面主题

  1. "[Theme]" — [N] tweets, [N] total likes Example: "@[handle]: [tweet excerpt]"
  2. "[Theme]" — [N] tweets
  1. "[主题]" — [N]条推文，总点赞量[N] 示例："@[handle]: [推文节选]"
  2. "[主题]" — [N]条推文

Most Engaged Tweets

互动量最高的推文

🔴 Most-liked negative: @[handle] ([N] likes): "[excerpt]" 🟢 Most-liked positive: @[handle] ([N] likes): "[excerpt]"
🔴 点赞最多的负面推文：@[handle]（[N]个点赞）："[节选]" 🟢 点赞最多的正面推文：@[handle]（[N]个点赞）："[节选]"

Volume Over Time

推文数量随时间变化

[Day 1]: [N] tweets | [Day 2]: [N] tweets | [Day 3]: [N] tweets
[第1天]：[N]条推文 | [第2天]：[N]条推文 | [第3天]：[N]条推文

Notable Spikes

显著峰值

[Date with highest volume] — [N] tweets | Likely cause: [describe if detectable from tweet context]
undefined
[推文数量最高的日期] — [N]条推文 | 可能原因：[根据推文上下文描述（如果可检测）]
undefined

Troubleshooting

故障排除

Sentiment feels inaccurate: Lexical analysis misses sarcasm. For high-stakes decisions, manually review the top 20 tweets per bucket. Topic too broad: Narrow the search term. "Apple" returns tech and food — use "Apple iPhone" instead. Very low tweet volume: Topic may not be actively discussed on Twitter right now. Expand date range.
情绪分析结果不准确： 词汇分析无法识别讽刺。对于高风险决策，请手动查看每个类别中排名前20的推文。 话题范围过广： 缩小搜索词范围。例如“Apple”会返回科技和食品相关内容——改用“Apple iPhone”。 推文数量极少： 该话题目前可能未在Twitter上被广泛讨论。扩大日期范围。