| Summary: | The implementation of sentiment analytical tools is largely centered between the use of lexicons resources and machine learning algorithms that are commonly used for general classification tasks. The results are commonly classified based on its polarities and are being engaged by a wide range of applications in the recent years. This report provides an insight of a comparison between the two approaches, and is compared with the hybrid approach which is developed in the report. The results are demonstrated on a competition task hosted by the International Workshop on Semantic Evaluation where the intensity of a different spectrum of emotions is evaluated. When run on the given datasets provided by the organisers, the results indicate that using lexicons resources alone performs poorly, especially when a general lexicon resource is selected. Traditional machine learning algorithms have been assessed as better candidates to produce better results however at the cost of expending more time to annotate the data. The hybrid approach which is the combination of both approaches has proven to yield better results with lesser costs involved than the implementation of individual approaches.
|
|---|