Some Facets of Argument Mining for Opinion Analysis

From AcaWiki
Jump to: navigation, search

Citation: Maria Garcia-Villalba, Patrick Saint-Dizier (2012) Some Facets of Argument Mining for Opinion Analysis. Proceeding of the 2012 conference on Computational Models of Argument: Proceedings of COMMA 2012 (RSS)
Internet Archive Scholar (search for fulltext): Some Facets of Argument Mining for Opinion Analysis
Wikidata (metadata): Q64514584
Download: fichier/COMMA-submission10.pdf
Tagged: Computer Science (RSS) opinion mining (RSS), online argumentation (RSS), argumentation mining (RSS)


Villalba & Saint-Dizier discuss the problem of argument extraction in opinion texts, in order to understand and present deeper analysis of why consumers like or dislike a product. Their goal is to induce a system of preferences from attribute-value pairs and a recommendation, identifying these from a review which contains opinion (aka "evaluative") expressions.

They make two main claims:

  1. "evaluative expressions with a 'heavy' semantic load" can be interpreted as arguments and
  2. "that the association of an evaluative expression with a discourse structure such as an elaboration, an illustration, or a reformulation must also be interpreted as an argument".

In addressing claim (1): They indicate that some, but not all, evaluative expressions function as arguments; they do not operationalize this, but merely provide descriptive examples. For instance: "an expression in the hotel domain such as: clean rooms has no real argumentative power because 'clean' is a standard value for rooms" In comparison, they say that "accueil familial (English gloss: you are welcomed as a family member)" incorporates an argument, since 'familial' has a rich semantic structure, with the probable interpretation: 'because the owners behave as if you were part of their family'. They describe this by saying that "'familial' is a higher-order adjective" -- yet this does not indicate how to determine or test which adjectives are "higher-order". They see the list of adjectives and modifiers and the semantic features underlying the adjectives (which they envision as extracted and combined) as an "informal set of arguments".

In addressing claim (2): They see the discourse structures accompanying evaluative expressions as "forms of explanation that realize argument schemas". But they do not construct the schemas.

They argue that, when combined with an evaluative expression, "elaborations, illustrations and other rhetorical relations related to explanation" acquire "argumentative power or orientation" by behaving "as argument supports and make explicit the semantic and pragmatic function of the support: e.g. they justify, illustrate, develop the evaluative expression".

They then detail, for various discourse relations, how they relate to the evaluations:

  • Justification clarifies the reasons for an evaluation

Some types of elaboration support the evaluation:

  • illustration
  • enumeration ("adds strength")
  • precision ("adds specific information")
  • comparison ("reinforce a statement, making a parallel with a better or worse situation or object, more or less explicit")

Other types of elaboration stress the importance of the attribute:

  • via consequence ("develops the advantages of a positive or a negative evaluation")
  • Contrast and Concession generally weaken the evaluation

Some discourse relations do not really have argumentative force, such as reformulation.