What it does
This KPI evaluates the emotional tone of text in marketing assets, including both written text and voiceover scripts. It uses a deep learning model trained to discern whether the text conveys positive, neutral, or negative sentiment.
Why it matters
The sentiment conveyed in text is crucial for emotional engagement. Positive sentiment strengthens emotional connections, builds brand loyalty, and encourages behaviors like purchases and sharing, while negative sentiment can have the opposite effect.
How it works
Text is either extracted from images through optical character recognition models or transcribed from videos viaspeech-to-text models. The sentiment analysis model then analyzes the text and assigns a sentiment score from 0 (negative) to 100 (positive). This score is contextualized against norms for different asset types to categorize the sentiment as low, medium, or high.
How to improve
-
Incorporate emotionally resonant and aspirational language to engage the audience.
-
Add sensory details to make the text more vivid and memorable.
-
Personalize the message and offer solutions to create a sense of relevance and utility.
-
Employ humor and empathy to establish a stronger emotional connection.
AI models used
The model was first trained to understand text in general, using a vast corpus of text from diverse sources, encompassing over 100 million words. It was then fine-tuned and trained to predict the sentiment of human-rated words (>20,000 words). The trained model has an accuracy of 93% in correctly predicting the human sentiment of texts.