Skip to content
Browse courses

NLP (Concepts & Pipeline)

Entity Sentiment Analysis

In shortNLP that scores the sentiment expressed towards each entity in a text, so one review can be positive about one thing and negative about another.

Updated 24 Oct 2026· 2 min read

Entity sentiment analysis combines entity extraction with sentiment analysis. Instead of giving a whole review one overall score, it finds each entity mentioned and scores how the writer feels about that entity specifically. This shows which parts of a product, service or brand people praise and which they complain about.

How it works

The model first identifies entities and every place they are mentioned, then scores the sentiment of the language around each mention and combines the results per entity. In the Google Cloud Natural Language API (the analyzeEntitySentiment method), each entity gets:

  • a score from -1.0 (negative) to 1.0 (positive), and
  • a magnitude from 0 upwards, showing how much emotional language there is overall.

IBM Watson NLU offers the same idea through sentiment options on its entities and keywords features, and Amazon Comprehend has targeted sentiment.

How it’s used in SEO and marketing

  • Mining customer reviews to see how people feel about delivery, pricing, sizing or customer service.
  • Comparing sentiment towards your brand and competitors in forums, reviews or social posts.
  • Checking how your brand is described in third-party content and AI-cited sources.
  • Turning feedback forms into a prioritised list of issues for product and CX teams.
  • Finding positive themes to reuse in landing page copy and FAQs.

Example

A review reads: “The jacket fits perfectly, but delivery took two weeks and customer service never replied.” Document-level sentiment might come out close to neutral because the praise and the complaints cancel out. Entity sentiment separates them: jacket positive, delivery negative, customer service negative. MLforSEO’s customer review lesson plots results like these on a chart, with sentiment on one axis and entity prominence or magnitude shown by size, so the biggest pain points stand out.

Limitations and things to watch

  • Sarcasm, comparisons (“better than X”) and negation still trip models up.
  • Entities may be split or merged inconsistently (for example “delivery” and “shipping”), so group them before reporting.
  • Language support for entity sentiment is narrower than for basic entity extraction; check the provider’s docs.
  • APIs disagree. In the independent test cited in MLforSEO’s API comparison, three sentiment APIs agreed with each other and with human labels on only 43.3% of items. Validate on a labelled sample first.

Learn it on MLforSEO

Where Entity Sentiment Analysis shows up on MLforSEO

Learn it in a course

Read about it on the blog

Try it yourself

Sources and further reading

Other terms from NLP (Concepts & Pipeline)

See all 17 NLP (Concepts & Pipeline) terms →