Opinion Mining — Aspect-Based Sentiment Extraction
Also known as: aspect-based sentiment analysis, opinion extraction, Görüş Madenciliği (Opinion Mining)
Opinion mining is a natural-language-processing task that systematically extracts and analyses user opinions about a product, service, or topic — identifying the specific features (aspects) being discussed, the sentiment expressed toward each, and the opinion holders. Consolidated by Bing Liu (2012), it goes beyond a single document-level label to produce structured aspect–opinion–holder records.
Key highlights
- Produces fine-grained, feature-level opinion rather than a single document-wide label.
- Identifies aspect–opinion–holder structure, capturing who said what about which feature.
- Turns unstructured review text into structured, comparable records that can be summarised across a corpus.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Opinion mining fits when you have text data containing opinions and you need feature-level rather than overall sentiment — for instance product or service reviews where different attributes are praised and criticised within the same text. A domain ontology or product feature list is not required but helps anchor aspect extraction. As a rule of thumb, gather at least around 50 opinion documents; with no text data the method cannot run.
Strengths & limitations
- Produces fine-grained, feature-level opinion rather than a single document-wide label.
- Identifies aspect–opinion–holder structure, capturing who said what about which feature.
- Turns unstructured review text into structured, comparable records that can be summarised across a corpus.
- More involved than document-level sentiment: aspects must be extracted before opinions can be attached to them.
- Accuracy improves when a domain ontology or product feature list is available to guide aspect extraction.
- Needs a reasonable volume of opinion text; very small corpora give unstable aspect-level summaries.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How is opinion mining different from sentiment analysis?
Document-level sentiment analysis assigns one polarity label to a whole piece of text. Opinion mining is finer-grained: it first extracts the specific aspects being discussed, then determines the sentiment toward each aspect and the opinion holder, producing structured aspect–opinion–holder records instead of a single verdict.
Do I need a product feature list or ontology?
It is not strictly required, but a domain ontology or product feature list helps anchor aspect extraction and improves accuracy by defining what counts as an aspect in your domain.
How much text do I need?
As a rule of thumb, gather at least around 50 opinion documents. Aspect-level summaries are unstable on very small corpora because each aspect may appear only a handful of times.
What does the output look like?
Rather than one label per document, the output is a set of aspect–opinion–holder triples — the feature discussed, the polarity expressed toward it, and where identifiable the holder — which can then be aggregated feature by feature across the corpus.
Sources
- 1.Liu, B. (2012). Sentiment Analysis and Opinion Mining. Morgan & Claypool.
- 2.Pang, B. & Lee, L. (2008). Opinion Mining and Sentiment Analysis. Foundations and Trends in Information Retrieval, 2(1-2), 1-135.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Opinion Mining. ScholarGate. https://scholargate.app/text-mining/opinion-mining