An integrated semi-automated framework for domain-based polarity words extraction from an unannotated non-English corpus

摘要

Building sentiment analysis resources is a fundamental step before developing any sentiment analysis model. Sentiment lexicons are one of these critical resources. However, many non-English languages suffer from a severe shortage of these resources and lexicons. This study proposes an integrated framework for extracting domain-based polarity words from unannotated massive non-English corpus. The framework consists of three layers, namely lexicon-based, corpus-based and human-based. The first two layers automatically recognize and extract new polarity words from a massive unannotated corpus using initial seed lexicons. A key advantage of the proposed framework is that it only needs an initial seed lexicon and unannotated corpus to start the extraction process. Therefore, the framework is semi-automated due to the use of seed lexicons. Experiments on three languages indicate the proposed framework outperformed existing lexicons, achieving F-scores of 77.8%, 83.8% and 68.6% for the Arabic, French and Malay lexicons, respectively.