Subjectivity in text expresses an individual's private state. This private state may be expressed through speculations, opinions, or evaluations. Each subjective expression can be associated with a subjective strength. Subjective strengths of clauses can be labeled as neutral, low, medium, high, or extreme. Subjective expressions can not only be described at the sentence level, but also can identified at different clause levels within the sentence.
For my ISSP 2030 project, I use Support Vector Machine (SVM) classifiers to learn the subjective strength at different clause levels given several different feature sets. The results obtained from the SVM classifiers will then be compared with those from BoosTexter. The SVM classifier tends to have a lower mean-squared error but a higher misclassification error.
This work was done in conjunction with Dr. Wiebe, Dr. Hwa, and Theresa Wilson.