Opinion Mining
Opinion mining is, in essence, another name for sentiment analysis, but it usually emphasizes the structured extraction of opinions: not only whether a text is positive or negative, but about what, by whom and when. We explain Bing Liu's quintuple model, its tasks and its relation to aspect-based analysis.
Opinion mining is the field that analyzes the opinions, appraisals and attitudes that people express in a text toward entities such as products, services, organizations or topics. It is worth clarifying up front: “opinion mining” and “sentiment analysis” are almost always used as synonyms—we refer to the sentiment analysis entry for the basics. The difference is one of emphasis: opinion mining usually highlights the structured extraction of the opinion, beyond classifying its polarity.
The quintuple model
The contribution that best captures that structured approach is that of the researcher Bing Liu, who models an opinion as a quintuple of five components: the entity being opined about, the specific aspect of that entity, the orientation of the sentiment (positive, negative or neutral), the holder of the opinion (who expresses it) and the time at which it is expressed. With those five data points, an opinion stops being a mere “like” and becomes exploitable information.
Its tasks
From that model derive the tasks of opinion mining: extracting the entities and their aspects, determining the orientation of the sentiment toward each aspect, and identifying the holder and the time. Its practical heart is aspect-based analysis, which distinguishes different opinions within a single sentence—“the camera is excellent, but the battery does not last long”—instead of giving a single label to the whole.
Summarizing and separating
Two complementary tasks round it off. Opinion summarization aggregates many appraisals about an entity and its aspects into a structured synthesis—what proportion of users praise the battery, how many criticize the screen—because a single opinion is not enough. And subjectivity detection separates what are verifiable facts from what are personal opinions, an essential preliminary step.
What it is for
Its applications are clear and much in demand: the analysis of product reviews, reputation management, market research and business intelligence, wherever understanding what many people think, and about what exactly, has value.
This article was produced with artificial intelligence under human editorial oversight.
