Feature Set Ensembles for Sentiment Analysis of Tweets

Griol Barres, David; Kanagal-Balakrishna, C.; Callejas Carrión, Zoraida

doi:https://doi.org/10.1007/978-3-030-51870-7_10

GriolKanagalCallejas21.pdf (482.6Ko)

Identificadores

URI: https://hdl.handle.net/10481/80255

DOI: https://doi.org/10.1007/978-3-030-51870-7_10

Exportar

Editorial

Springer

Materia

Sentiment analysis

Twitter

Date

2021

Referencia bibliográfica

This is a pre-print version of the chapter: Griol, D., Kanagal-Balakrishna, C., Callejas, Z. (2021). Feature Set Ensembles for Sentiment Analysis of Tweets. In: Phillips-Wren, G., Esposito, A., Jain, L.C. (eds) Advances in Data Science: Methodologies and Applications. Intelligent Systems Reference Library, vol 189. Springer, Cham. https://doi.org/10.1007/978-3-030-51870-7_10 (https://link.springer.com/chapter/10. 1007/978-3-030-51870-7_10)

Résumé

In recent years, sentiment analysis has attracted a lot of research attention due to the explosive growth of online social media usage and the abundant user data they generate. Twitter is one of the most popular online social networks and a microblogging platform where users share their thoughts and opinions on various topics. Twitter enforces a character limit on tweets, which makes users find creative ways to express themselves using acronyms, abbreviations, emoticons, etc. Additionally, communication on Twitter does not always follow standard grammar or spelling rules. These peculiarities can be used as features for performing sentiment classification of tweets. In this chapter, we propose a Maximum Entropy classifier that uses an ensemble of feature sets that encompass opinion lexicons, n-grams and word clusters to boost the performance of the sentiment classifier. We also demonstrate that using several opinion lexicons as feature sets provides a better performance than using just one, at the same time as adding word cluster information enriches the feature space.