How to assess customer opinions beyond language barriers?

Author

Denecke, Kerstin

Author_Institution

Res. Center L3S, Hannover

fYear

2008

fDate

13-16 Nov. 2008

Firstpage

430

Lastpage

435

Abstract

The main focus of this paper is to introduce an approach to sentiment classification for documents in different languages. The method is based on language-specific resources available for English. First, documents are translated to English using standard translation software. For polarity detection, sentiment-bearing terms are identified by means of SentiWordNet. Polarity scores calculated for words of three word classes are exploited by a machine learning classifier for determining the document polarity. The introduced method is tested and evaluated on movie reviews in six different languages. The results show that polarity can be correctly determined even if language specific resources are unavailable.

Keywords

document handling; language translation; learning (artificial intelligence); natural languages; pattern classification; SentiWordNet; customer opinions; document polarity; language barriers; language-specific resources; machine learning classifier; sentiment document classification; sentiment-bearing terms; standard translation software; Discussion forums; Frequency; Information analysis; Internet; Machine learning; Motion pictures; Natural languages; Performance analysis; Software standards; System testing;

fLanguage

English

Publisher

ieee

Conference_Titel

Digital Information Management, 2008. ICDIM 2008. Third International Conference on

Conference_Location

London

Print_ISBN

978-1-4244-2916-5

Electronic_ISBN

978-1-4244-2917-2

Type

conf

DOI

10.1109/ICDIM.2008.4746812

Filename

4746812