Document-Level Sentiment Classification: An Empirical Comparison Between Using SentiWordNet Lexicon and Using SentiTFIDF

• Main Authors: Meng-Wen Lee, 李孟紋
• Other Authors: Shing-Hwang Doong
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/k7646v

碩士 === 樹德科技大學 === 資訊管理系碩士班 === 103 === In the field of sentiment analysis, SentiWordNet published in 2006 has been a very important lexical resource for document preprocessing. Through SentiWordNet we can find the strength of emotional words in a document. Ghag and Shah (2014) proposed a SentiTFIDF algorithm to classify the sentiment of movie reviews, and they obtained good results in their experiments. This thesis compares the differences between SentiWordNet with support vector machine (SVM) and the SentiTFIDF algorithm in a classification task of the sentiment of movie reviews. After conducting extensive experiments, we conclude as follows: (1) SentiTFIDF is better than SentiWordNet with SVM in accuracy; (2) SentiWordNet with SVM is better than SentiTFIDF in prediction stability; (3) SentiTFIDF is more complicated than SentiWordNet with SVM in the experimental procedure; (4) SentiTFIDF is more time-consuming than SentiWordNet with SVM.