This paper compares two types of sentiment analysis (SA), global sentiments and the opinions of people regarding a specific target topic. Document-level SA can decide the polarity of a document at a global level, while target-level SA determines the opinions toward a given target word. The authors used social media as the platform for this analysis and assigned polarity labels to textual elements. For both areas, the authors used a database consisting of Twitter messages and product reviews. In general, their task was to determine if the overall sentiment of the textual element is positive or negative.
The system used in this research was based on supervised machine learning to classify documents into polarity classes. The document-level analysis required preprocessing to remove special characters and emoticons, which enabled the polarity of the text to be characterized. The target-level SA used “a syntactic parser to select clauses that are related to the target ... and used these clauses to detect sentiments.”
In conclusion, according to the authors, they were able “to distinguish positive and negative sentiments better” than previous SA methodologies. However, as mentioned in the final paragraph, this accuracy is highly dependent on several factors like the text language, length, and genre. This paper would be of interest to academics and people who work in the field of SA.