| Summary: | 碩士 === 國立高雄應用科技大學 === 資訊工程系 === 106 === With the rapid development of the worldwide computer networks, the internet has long been deeply rooted in daily life. For instance, after the emergence of social media, more and more people chose to express their personal feelings and opinions online. So that the sentiments in the posts on social media are closer to people’s real sentiments. Social media posts, therefore, can be considered a good corpus for sentiment analysis. The main purpose of this study is to predict the sentimental polarities of social media posts via measuring syntactic similarity. Posts were collected mainly from five bulletin boards with apparent sentimental polarities in the social media PTT forum, namely, the Happy and Lucky bulletin boards with positive sentimental polarity and the Hate, Broken-heart, and Sad bulletin boards with negative sentimental polarity.
There are two research methods in this study. One is to use the n-gram model to analyze the syntactic structures and try to identify syntactic patterns that can represent either positive or negative sentiment. Another method is to use the partial tree kernel to calculate the syntactic similarity of sentences and, at the same time, use a support vector machine to establish a predictive model to analyze the sentiment polarities. Finally, the models separately established by the two methods above that can recognize sentimental polarities of social media posts are used to predict the sentimental polarities of social media posts.
Compared with using words or shallow syntactic structures to predict the sentimental polarities of posts in previous research, the entire parsing tree was directly analyzed, and the syntactic structures with positive or negative sentiment polarity were found in our research. The experimental results indicate that the method used showed outstanding performance in predicting sentimental polarities of social media posts.
|
|---|
