Sentiment classification of long newspaper articles based on automatically generated thesaurus with various semantic relationships

摘要

The paper describes a new approach for sentiment classification of long texts from newspapers using an automatically generated thesaurus. An important part of the proposed approach is specialized thesaurus creation and computation of term's sentiment polarities based on relationships between terms. The approach's efficiency has been proved on a corpus of articles about American immigrants. The experiments showed that the automatically created thesaurus provides better classification quality than manual ones, and generally for this task our approach outperforms existing ones.