an article by Habib Ullah Khan (Qatar University, Doha) and Duncan Peacock (University of Liverpool, UK; University of Sheffield, UK) published in International Journal of Work Organisation and Emotion Volume 10 Number 2 (2019)
Abstract
Work organisations are increasingly interested in using sentiment analysis algorithms to get rapid feedback from microblogging platforms such as Twitter. However, real-life posts can differ from the training data. The subject domain may vary and/or emojis and emoticons used to clarify, enhance or even reverse the sentiment of a post.
This paper studies the effect of emojis, emoticons and subject on polarity classification using nine tweet-related sentiment analysis web services.
A web application was developed to extract from the live Twitter stream, and twelve specific research test sets were created.
These were labelled by volunteers, uploaded back into the application and then compared against nine different sentiment analysis web services using two- and three-class accuracy measures. Distinct differences were found in the performance of the sentiment analysis web services of organisations.
Sentiment analysis web services can vary significantly in classification performance depending on the effect of emoticons and emojis.
Labels:
Twitter, sentiment_analysis, polarity_classification, web_services, emoticons, emojis, comparison, benchmark,
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment