A New Approach for A Domain-Independent Turkish Sentiment Seed Lexicon Compilation
Ekin Ekinci and Sevinç
Omurca
Department of Computer Engineering, Kocaeli University,
Turkey
Abstract: Sentiment
analysis deals with opinions in documents and relies on sentiment lexicons;
however, Turkish is one of the poorest languages in regard to having such
ready-to-use sentiment lexicons. In this article, we propose a
domain-independent Turkish sentiment seed lexicon, which is extended from an
initial seed lexicon, consisting of 62 positive/negative seeds. The lexicon is
completed by using the beam search method to propagate the sentiment values of
initial seeds by exploiting synonym and antonym relations in the Turkish
Semantic Relations Dataset. Consequently, the proposed method assigned 94 words
as positive sentiments and 95 words as negative sentiments. To test the
correctness of the sentiment seeds and their values the first sense, the total
sum and weighted sum algorithms, which are based on SentiWordNet and SenticNet
3, are used. According to the weighted sum, experimental results indicate that
the beam search algorithm is a good alternative to automatic construction of a
domain-independent sentiment seed lexicon.
Keywords: Sentiment lexicon, beam search, pattern generation, turkish
language, unsupervised framework.