What is sentiment in MosaicVoice?
Updated
What is sentiment in MosaicVoice?
Sentiment is a numerical score that represents the emotional tone of a customer during a call.
The scoring system
Score Range: -1.0 (most negative) to +1.0 (most positive)
- Positive: Score of 0.3 or higher
- Neutral: Score between -0.3 and 0.3
- Negative: Score of -0.3 or lower
How sentiment is determined
The model uses three main signals from the transcript:
a. Word Choice and Polarity
It looks for emotionally loaded words or phrases.
- Positive cues: “thank you so much,” “that’s great,” “I appreciate it”
- Negative cues: “this is ridiculous,” “I’m upset,” “terrible service”
Example: Customer: “I’m so frustrated, you’ve been charging me twice every month.” → Score: -0.7, Label: "negative"
b. Tone and Politeness
Even neutral words can carry sentiment if framed politely or rudely.
- “Could you please check my account?” → more positive/neutral.
- “Why do I have to keep repeating myself?” → more negative.
c. Contextual Flow
The model weighs overall tone across the call, not just isolated words.
- If a call starts negative but ends positively after resolution, the final sentiment may lean positive or neutral depending on weight of resolution vs. complaint.
Example: Start: “I’m angry, this charge is wrong.” (-0.6) End: “Thanks, I’m glad you fixed it.” (+0.7) → Final averaged/weighted sentiment might be around +0.2 → "neutral".
Key Points to Remember
- Sentiment is calculated after the call ends, not in real-time
- Each call gets one overall sentiment score, not segment-by-segment analysis
- The score is stored and can be used for reporting and analysis
- Multiple factors beyond the AI score can flag a call as negative (profanity, specific emotions)
Technical Note
The sentiment feature can be enabled or disabled per customer through the is_sentiment_active configuration setting.