Sentiment Analysis

Using AI to identify whether language expresses positive, negative, neutral, or mixed attitude.

One review, two different opinions

Read this fictional visitor review: “The prints were wonderful, but the booking form froze twice.”

Exhibition · positive

“Wonderful” evaluates the prints. A curator looking at visitor reactions would want to preserve that praise.

Booking form · negative

“Froze twice” describes a problem with the booking experience. A digital services team needs the complaint and the wording behind it.

What gets lost in a single label?

A document-level label such as “mixed” can summarize the review, while an aspect-level reading identifies what needs attention. Keep the relevant span and target together. Ask for human review when sarcasm, quotation, or unclear references make the attitude uncertain.

How Sentiment Analysis Works

Older sentiment systems relied heavily on word lists and rules. Modern systems often use natural language processing, transformers, and sometimes large language models to interpret tone in fuller context. This helps with phrases where the surface wording is misleading, such as sarcasm, negation, or domain-specific language.

Some systems classify sentiment into simple categories, while others use aspect-based sentiment analysis to score specific features or targets, estimate finer-grained emotions, or track shifts over time. In voice workflows, sentiment analysis may also combine text transcripts with acoustic cues, making it part of a broader multimodal learning pipeline.

Follow the idea

Natural language processing supplies methods for analyzing the text.

Social listening groups public mentions for broader review.

Related concepts (10)
Related Yenra articles (17)

Source · Microsoft documentation on sentiment analysis and opinion mining

Return to Sentiment Analysis in the A–Z glossary