OpinionMining-ML

In this paper we propose OpinionMining-ML, a new XML-based formalism for tagging textual expressions conveying opinions on objects that are considered relevant in the state of affairs. The need of such a formalism is motivated by the lack of standards for Opinion Mining (a.k.a. Sentiment Analysis) that obey to certain requirements of efficiency, ease of manual annotation, scalability, and, most of all, that aim at satisfying the real goal of Sentiment Analysis applications. Opinion Mining is an Information Retrieval task, so that its output should be designed for being usable and fruitful from the perspective of a search engine. Our contribution is twofold. First, we present a standard methodology for the annotation of affective statements in text that is strictly independent from any application domain. The second and orthogonal part of the approach regards instead the domain-specific adaptation that relies on the use of an ontology of support, that is domain-dependent by definition. We finally evaluate our proposal by means of fine-grained analyses of the disagreement between different annotators.

► It introduces a formalism for the annotation of sentiments and opinions in texts.
► OpinionMining-ML has been thought to be application-oriented and easy to use.
► The paper presents a text corpus annotated according to OpinionMining-ML.
► It demonstrates its validity through an inner-annotation agreement evaluation.
► It proposes several extensions that may cover complex cases.