Conference Paper

Sentence and Expression Level Annotation of Opinions in User-Generated Discourse

Conference: Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics

ABSTRACT In this paper, we introduce a corpus of consumer reviews from the rateitall and the eopinions websites annotated with opinion-related information. We present a two-level annotation scheme. In the first stage, the reviews are analyzed at the sentence level for (i) relevancy to a given topic, and (ii) expressing an evaluation about the topic. In the second stage, on-topic sentences containing evaluations about the topic are further investigated at the expression level for pinpointing the properties (semantic orientation, intensity), and the functional components of the evaluations (opinion terms, targets and holders). We discuss the annotation scheme, the inter-annotator agreement for different subtasks and our observations.

Download full-text


Available from: Niklas Jakob, Aug 18, 2015
1 Follower
  • Source
    • "In the last years, several corpora have been annotated with information related to modality and polarity, which have made it possible to develop machine learning systems. Annotation has been performed at different levels: word (Hassan and Radev, 2010), expression (Baker et al., 2010; Toprak et al., 2010), sentence (Medlock and Briscoe, 2007), event (Saurí and Pustejovsky, 2009), discourse relation (Prasad et al., 2006), text (Amancio et al., 2010), and scope of negation and modality cues (Vincze et al., 2008). Thanks to the existence of the BioScope corpus, the scope processing task was recently introduced. "
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper we summarize existing work on the recently introduced task of processing the scope of negation and modality cues; we analyse the scope model that existing systems can process, which is mainly the model reflected in the annotations of the biomedical corpus on which the systems have been trained; and we point out aspects of the scope finding task that would be different based on observations from a corpus from a different domain and nature.
  • Source
    • "This much harms the detection performance due to the noise in the automatically-generated training data. Note that Toprak et al. [9] presents a manually annotated corpus which considers the opinion expression in the sentence level from many sides, such as polarity, strength, modifier, holder, and target. The concept of polarity shifting is also mentioned. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Polarity shifting has been a challenge to automatic sentiment classification. In this paper, we create a corpus which consists of polarity-shifted sentences in various kinds of product reviews. In the corpus, both the sentimental words and shifting trigger words are annotated. Furthermore, we analyze all the polarity shifted sentences and categorize them into five categories: opinion-itself, holder, target, time and hypothesis. Experimental study shows the agreement of annotation and the distribution of the five categories of polarity shifting.
  • Source
    Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing; 01/2010
Show more