Skip to main content
Glossary · Techniques

Aspect-Based Sentiment Analysis.

Sentiment scored at the attribute level — battery, screen, weight — not averaged across the whole review.

Definition

Aspect-based sentiment analysis (ABSA) is a natural-language processing technique that identifies the specific product attributes or aspects mentioned in a piece of feedback and scores sentiment separately for each one. Instead of classifying a full review as positive or negative, ABSA returns a set of (aspect, sentiment) pairs — for example, battery: positive, weight: negative, screen: neutral — reflecting how reviewers actually write.

Definition

Aspect-based sentiment analysis — ABSA — is the task of extracting, from a span of text, the product attributes the writer referenced and the polarity (positive, negative, neutral, mixed) of each. A four-star review of a laptop that says "keyboard is fantastic, battery dies in four hours, screen is average" produces three rows: keyboard positive, battery negative, screen neutral. Document-level sentiment loses all of that; it would return something like "mildly positive, 3.6/5."

Modern ABSA implementations combine a taxonomy of aspects (either industry-specific or learned from the corpus) with a classifier that operates at sentence or clause level. Good implementations also handle negation, scope, and implicit aspects ("it's a brick" implies weight). A strong aspect taxonomy for a product category is typically 40–120 terms; too short and real complaints get bucketed under "other," too long and the signal fragments.

Why it matters

Consumer product reviews are compound by nature. Shoppers list what they like, what they don't, and what they're ambivalent about — often in the same sentence. Whole-review sentiment collapses that into a single score that is simultaneously true and useless. A product team reading "70% positive" learns nothing actionable; a product team reading "screen sentiment dropped 18 points on this SKU while keyboard sentiment held" knows exactly which component to investigate.

ABSA is the technique that makes sentiment actually usable for product, QA, and consumer-insights work on physical goods, because every physical good is a bundle of attributes that can be praised and criticized independently. It is also what lets a brand compare two SKUs in the same category on the attributes that matter — rather than on an undifferentiated star average that reflects everything and nothing.

Example

A consumer-electronics brand monitors a flagship headphone SKU on Amazon, Best Buy, and Bazaarvoice-syndicated retailer pages. The average rating is 4.3 stars and holding. Document-level sentiment reads "positive, stable." ABSA tells a different story: "comfort" sentiment is at +82, "sound" at +71, "charging case" at +12, and "Bluetooth pairing" at -34 and worsening over the last 30 days.

The product team pulls the Bluetooth pairing reviews — 217 of them across the three channels — finds they cluster on a recent firmware release, and ships a patch. Average rating never moved enough to flag. Aspect-level sentiment flagged it in week one. The comparable prior-generation SKU, by contrast, shows stable "Bluetooth pairing" sentiment, which tells the QA team the regression is firmware-specific and not a hardware issue.

Ask Indellia

Have a specific question?

Indellia's AI agents answer with citations from real customer feedback across Amazon, Walmart, Best Buy, and 20+ retail channels.

Get started

Score sentiment where it actually moves.

Indellia runs aspect-level sentiment on every review, ticket, and survey response across 20+ retail and review channels, linked to the SKU it describes. Unlimited users. Unmetered data. $495/mo SME, $1,995/mo Mid-Market.