ABSA & Sentiment

Flat Sentiment Is Lying to You: Why Aspect-Level Analysis Changes Everything

One number, six owners, zero actions.

Marc DuboisPrincipal Data Scientist, InsightNarrator30 June 202610 min read

The difference between “sentiment is +0.3” and “Support is -0.5 but Product is +0.7.”

Averaging is the bug

Document-level sentiment averages opposing opinions inside a single sentence. The result is a number that is technically correct and operationally useless.

Aspects give the score an owner

Aspect extraction identifies what the customer is talking about, then scores each target separately. Suddenly the analysis routes: Product to the roadmap, Support to operations, Billing to finance.

Want to try this on your own data? Run a free VOC health check — no credit card, EU-processed.

Start free

Key takeaways

Never report a single sentiment score for a mixed-opinion corpus.
An aspect is only useful if a team owns it.

About the author

Marc Dubois

Principal Data Scientist, InsightNarrator

Builds the multi-agent analysis pipeline and the taxonomy governance layer. Formerly NLP research in telco churn modelling.

Connect on LinkedIn

Keep reading

CX Strategy

Why NPS Is a Number and Your Verbatims Are the Story — And What to Do About It

Your NPS is +32. Now what? Most teams stop at the score. But the answers to “why did they rate us this way?” are in the verbatims — and almost nobody analyses them at depth. Here's how five analysis techniques transform NPS from a metric into an action plan.

11 min read

CX Strategy

The 43 Analysis Techniques Every CX Team Should Know

A field guide to the 43 analysis types in the InsightNarrator catalogue — and which ones your CX team actually needs first.

12 min read

Data Sovereignty

The Hidden Risk of US-Hosted CX Tools Under the CLOUD Act

Why your customer feedback platform might be creating a GDPR compliance gap you didn't know about.

8 min read

Start free trial

14 days, no credit card, EU-processed.

Start free trial Book a demo