Combining sentiment analysis scores to improve accuracy of polarity classification in MOOC posts

Laroca, Herbert; Rocio, Vitor

http://hdl.handle.net/10400.2/8703

Use this identifier to reference this record.

Name:	Description:	Size:	Format:
artigo_epia_revisado_20190617.pdf		240.02 KB	Adobe PDF	Download

Send Feedback

Authors

Laroca, Herbert

Rocio, Vitor

Abstract(s)

Sentiment analysis is a set of techniques that deal with the verification of sentiment and emotions in written texts. This introductory work aims to explore the combination of scores and polarities of sentiments (positive, neutral and negative) provided by different sentiment analysis tools. The goal is to generate a final score and its respective polarity from the normalization and arithmetic average scores given by those tools that provide a minimum of reliability. The texts analyzed to test our hypotheses were obtained from forum posts from participants in a massive open online course (MOOC) offered by Universidade Aberta de Portugal, and were submitted to four online service APIs offering sentiment analysis: Amazon Comprehend, Google Natural Language, IBM Watson Natural Language Understanding, and Microsoft Text Analytics. The initial results are encouraging, suggesting that the average score is a valid way to increase the accuracy of the predictions from different sentiment analyzers.