akrindev/sentiment-analyzer

Sentiment Analyzer

This skill enables AI agents to analyze sentiment and emotion in video content.

When to Use

• User wants to identify emotional peaks in content
• Detecting positive/negative/neutral sentiment
• Finding reaction moments (surprise, excitement, anger)
• Analyzing the emotional journey throughout the video
• Creating highlight clips from emotional moments

Detection Methods

1. Transcript-Based Keyword Detection

Analyzes transcript for emotion keywords:

Positive: excited, amazing, incredible, wow, love, happy, joy Negative: terrible, awful, hate, sad, angry, frustrated, disappointing Surprise: oh my god, what!, unbelievable, shocked, can't believe Enthusiasm: let's go, come on, yes!, awesome, fantastic

2. AI-Based Emotion Detection (Gemini)

Uses Gemini API for advanced emotion analysis:

• Context-aware sentiment detection
• Emotion classification (happy, sad, angry, surprised, neutral)
• Intensity scoring
• Speaker emotion tracking

3. Audio Feature-Based Detection

Analyzes audio characteristics:

• Pitch variations
• Volume/loudness changes
• Speech rate changes
• Energy patterns

Available Scripts

scripts/analyze_sentiment.py

Analyze sentiment in video content.

Usage:

python skills/sentiment-analyzer/scripts/analyze_sentiment.py <video_path> [options]

Options:

• --method: Analysis method (keywords, ai, audio) - default: keywords
• --transcript-path: Path to transcript SRT/VTT file (for keyword detection)
• --output, -o: Output JSON path (default: <video_path>_sentiment.json)
• --window-size: Analysis window size in seconds - default: 5.0

Examples:

Analyze from transcript:

python skills/sentiment-analyzer/scripts/analyze_sentiment.py video.mp4 --transcript-path video.srt

Analyze with Gemini AI:

python skills/sentiment-analyzer/scripts/analyze_sentiment.py video.mp4 --method ai

scripts/find_emotional_peaks.py

Find emotional peaks and significant moments.

Usage:

python skills/sentiment-analyzer/scripts/find_emotional_peaks.py <video_path> [options]

Options:

• --threshold: Peak detection threshold (0.0-1.0) - default: 0.7
• --emotion-type: Filter by emotion type (positive, negative, surprise, all) - default: all
• --output, -o: Output JSON path

Example:

python skills/sentiment-analyzer/scripts/find_emotional_peaks.py video.mp4 --threshold 0.8 --emotion-type positive

Output Format

{
  "video_path": "video.mp4",
  "method": "keywords",
  "overall_sentiment": {
    "positive": 0.45,
    "negative": 0.15,
    "neutral": 0.40
  },
  "emotional_peaks": [
    {
      "timestamp": 23.5,
      "duration": 3.2,
      "emotion": "positive",
      "intensity": 0.85,
      "text": "This is absolutely incredible!",
      "keywords": ["incredible"]
    },
    {
      "timestamp": 67.0,
      "duration": 2.8,
      "emotion": "surprise",
      "intensity": 0.90,
      "text": "Oh my god, I can't believe this!",
      "keywords": ["oh my god", "can't believe"]
    }
  ],
  "sentiment_timeline": [
    {
      "start": 0.0,
      "end": 30.0,
      "dominant_emotion": "neutral",
      "intensity": 0.4
    },
    {
      "start": 30.0,
      "end": 60.0,
      "dominant_emotion": "positive",
      "intensity": 0.7
    }
  ]
}

Emotion Keywords

Positive Emotions

• amazing, incredible, fantastic, awesome, wonderful, great, excellent
• love, happy, joy, excited, thrilled, delighted, pleased
• perfect, brilliant, outstanding, impressive, remarkable

Negative Emotions

• terrible, awful, horrible, bad, terrible, disappointing
• hate, angry, frustrated, annoyed, upset, sad, depressed
• worst, disgusting, pathetic, useless, failed

Surprise/Excitement

• wow, oh my god, what!, unbelievable, shocked, stunned
• surprised, amazed, astonished, incredible, no way
• excited, thrilled, pumped, fired up, let's go

Neutral/Calm

• okay, alright, fine, good, normal, regular, standard

Integration with Other Skills

After sentiment analysis, you can use these skills:

• highlight-scanner: Combine sentiment with other signals
• video-trimmer: Create clips from emotional peaks
• autocut-shorts: Full workflow for creating short clips

Common Workflow

1. User provides video file
2. Transcribe using video-transcriber
3. Analyze sentiment using this skill
4. Find emotional peaks
5. Create short clips from emotional moments

Tips

• Emotional peaks are excellent for viral content
• Positive + surprise emotions have highest viral potential
• Combine with laughter detection for even better results
• Consider surrounding context (3-5 seconds before/after)
• High-intensity emotions (>0.8) are premium clip candidates

References

• Sentiment analysis research papers
• Emotion detection in NLP
• Audio emotion analysis techniques