Sentiment analysis

Marketing Analytics for Business

Sarah DeAtley

Principal Data Scientist

Sentiment analysis basics

image showing happy, neutral, and sad face with checkboxes

Brands advertise on social media platforms like Twitter, Instagram, and Facebook
One way to monitor social media health is through sentiment analysis

Sentiment analysis: technique to determine if data is positive, negative, or neutral

¹ athree23 from Pixabay

Sentiment is subjective! Brands want to monitor in real-time
Context is key:
- "Well done" in sports has a different meaning than in restaurants

Sentiment can be more than positive, negative, or neutral:
- Angry, happy, or sad
- Urgent or non-urgent
Also consider emojis and sarcasm

image of steak on the left and golfers on the right with text "well done"

various emojis with label "reactions"

¹ Denis Cherkashin on Unsplash

Evaluate level of complexity and data size:
- Automatic for large data sets with multiple languages
- Manual for ease of interpretation with smaller data
Manual rules can train automatic NLP models (hybrid approach)

man scratching head looking at paper with happy, neutral, and smiling faces

¹ mohamed Hassan from Pixabay

various charts in a dashboard

NLP model options:
- Out-of-the-box model with NLP sentiment
- Build and maintain in-house NLP sentiment model
Still learning about NLP? Start with out-of-the-box model
Experience training NLP models? In-house has more flexibility
Consider long term reporting needs with both options!

¹ Carlos Muza on Unsplash

Marketing Analytics for Business