Automate Tweet Sentiment Analysis with n8n

Step 1: Trigger Apify Actor to Scrape Tweets We started with an HTTP Request node in n8n to run the Apify Tweet Scraper Actor. This retrieves tweets matching a specific search query. HTTP Request Configuration:

• Method: POST
• URL: https://api.apify.com/v2/acts/kaitoeasyapi~twitter-x-data-tweet-scraper-pay-per-result-cheapest/runs?token=YOUR_APIFY_TOKEN
• Authentication: Predefined Credential (Apify API)
• Headers:

  [
    { "Content-Type": "application/json" }
  ]

• Body (JSON):

  {
    "from": "realDonaldTrump",
    "twitterContent": "war OR conflict OR Iran OR Israel OR military OR missile OR attack OR sanctions OR diplomacy OR peace OR nuclear OR troop OR battlefield OR strike OR invasion OR ceasefire OR missilestrike OR defense OR escalation",
    "maxItems": 500,
    "queryType": "Top",
    "lang": "en"
  }

💡 Tip:

• You can customize twitterContent to match your niche. Example: “AI OR ChatGPT OR LLM OR machine learning” for AI trends, or “Olympics OR gold medal OR Paris 2024” for sports updates.
• You can also change the from field to scrape tweets from any person you want (e.g., kanyewest).
• You can change the queryType to any supported type, such as latest, top, or mixed depending on your needs.

Step 2: Extract Dataset ID from Apify Actor Response Apify returns a JSON object containing a datasetId after the actor runs. We captured this ID using an n8n Set or Function node for use in the next step.

Step 3: Fetch Tweet Data from Apify Dataset API We used another HTTP Request node to get the actual scraped tweets. HTTP Request Configuration:

• Method: GET
• URL: https://api.apify.com/v2/datasets/{{ $json["data"]["defaultDatasetId"] }}/items?token=YOUR_API_TOKEN&clean=true
• Authentication: None (API token included in URL)
• Headers: None
• Body: None

Step 4: Assign Unique IDs to Tweets To avoid duplicates and make later processing easier, we used a Code node to assign random unique IDs when missing.

const items = $input.all();
const transformedItems = items.map(item => {
  const { id: originalId, ...rest } = item.json;
  const numericId = Number(originalId);
  const newId = numericId === -1 ? Math.floor(Math.random() * 1000000000) : numericId;
  return { id: newId, ...rest };
});
return transformedItems;

Sometimes, if the actor doesn’t find real tweets (due to errors or no matches), it sends a mock tweet with id = -1. This code handles that by replacing -1 with a random unique ID.

You can also modify this node to filter out such mock tweets completely and output nothing if you prefer. If you don’t want to handle IDs manually, you can simply skip this step.

Step 5: Store Raw Tweets in Google Sheets We used the Google Sheets node to append the fetched tweets into our document. Configuration:

• Credential: Connected Google Sheet account
• Resource: Sheet Within Document
• Operation: Append or Update Row
• Document: Twitter
• Sheet: tweets
• Mapping: Mapped each tweet field manually to columns

Step 6: Run Sentiment Analysis on Tweets For sentiment analysis, we used Hugging Face’s cardiffnlp/twitter-roberta-base-sentiment model via an HTTP Request node. HTTP Request Configuration:

• Method: POST
• URL: https://api-inference.huggingface.co/models/cardiffnlp/twitter-roberta-base-sentiment
• Headers:

  [
    { "Authorization": "Bearer YOUR_HF_API_KEY" },
    { "Content-Type": "application/json" }
  ]

• Body (JSON):

  { "inputs": "{{ $json.text }}" }

💡 Tip: You can try changing the model to another one, like distilbert-base-uncased-finetuned-sst-2-english, for faster results while maintaining good accuracy. Feel free to explore other Hugging Face models to find the best fit for your needs!

Step 7: Map Hugging Face Labels to Human-Readable Sentiment : The Hugging Face sentiment model returns labels as LABEL_0, LABEL_1, and LABEL_2, which correspond to negative, neutral, and positive sentiments respectively. To convert these into more understandable terms, we used an n8n Code node with the following JavaScript code:

// Process all input items
const items = $input.all().map(item => {
  // Create a copy of the original JSON data
  const newItem = {...item.json};
  
  // Calculate sentiment if not already present
  if (!newItem.sentiment) {
    const labelMap = {
      "LABEL_0": "negative",
      "LABEL_1": "neutral",
      "LABEL_2": "positive"
    };
    
    // Get all available results
    const results = [];
    ['0', '1', '2'].forEach(key => {
      if (newItem[key]) {
        results.push(newItem[key]);
      }
    });
    
    // Determine sentiment if we have results
    if (results.length > 0) {
      const top = results.reduce((max, current) => 
        (current.score > max.score) ? current : max, 
        {score: -Infinity}
      );
      newItem.sentiment = labelMap[top.label] || 'unknown';
    } else {
      newItem.sentiment = 'unknown';
    }
  }
  
  return {json: newItem};
});

return items;

This Code node reads each item’s label, maps it to a clear sentiment category, and adds a new sentiment field with the result.