Performance Comparison of different Lexicons for Sentiment Analysis in Arabic

Awwad H., ALPKOÇAK A.

3rd European Network Intelligence Conference (ENIC), Wroclaw, Polonya, 5 - 07 Eylül 2016, ss.127-133, (Tam Metin Bildiri) identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası:
  • Doi Numarası: 10.1109/enic.2016.25
  • Basıldığı Şehir: Wroclaw
  • Basıldığı Ülke: Polonya
  • Sayfa Sayıları: ss.127-133
  • Anahtar Kelimeler: Sentiment Analysis, polarity classification, lexicon-based, Arabic, sentence-level, document-level
  • Dokuz Eylül Üniversitesi Adresli: Evet

Özet

This paper deals with the lexicon-based approach in document-level and sentence e-level sentiment analysis (SA) in Arabic. We experimented four different lexicons; a translation of Harvard IV-4 Dictionary (HarvardA), translation of the MPQA subjectivity lexicon developed by Pittsburgh University (HRMA) and two different implementation of MPQA. We evaluated all four lexicons with three datasets from different domains; one of them is about health comments (PatientJo), the second is from Twitter data, and the third is about books reviews (LABR). For sentence-level SA, we suggested six different methods for sentiment values and document polarity. The results show that the HRMA lexicon performs better than other lexicons in LABR while HarvardA perform better in PatientJo dataset. The results show that lexicon-based approach for document-level and sentence-level methods produce similar performance. We observed that giving extra weight for the first and last sentences in sentence-level approach improves the overall performance in terms of accuracy.