Loading...
Paper Type
Complete
Paper Number
1462
Description
The accuracy of different cloud-based technologies for sentiment analysis may vary based on attributes such as the length of the analyzed texts and the dominant sentiment in a corpus. A potential strategy to reduce the variability in accuracy is to create ensemble models formed by individual technologies. Our goal in this paper is to study the performance of different ensembles of cloud-based technologies for sentiment analysis. Overall, we find that ensemble models perform better on long texts, a scenario where individual technologies tend to struggle. We also find that score-based ensembles perform better than label-based ensembles. Besides being of value to practitioners, we discuss how our results might increase the reliability of research findings that rely on sentiment analysis. In particular, we argue that ensemble models may reduce the chances of sentiment-dependent results being driven by a single technology.
Recommended Citation
Carvalho, Arthur and Xu, Jiaozhe, "Studies on the Accuracy of Ensembles of Cloud-Based Technologies for Sentiment Analysis" (2021). AMCIS 2021 Proceedings. 12.
https://aisel.aisnet.org/amcis2021/art_intel_sem_tech_intelligent_systems/art_intel_sem_tech_intelligent_systems/12
Studies on the Accuracy of Ensembles of Cloud-Based Technologies for Sentiment Analysis
The accuracy of different cloud-based technologies for sentiment analysis may vary based on attributes such as the length of the analyzed texts and the dominant sentiment in a corpus. A potential strategy to reduce the variability in accuracy is to create ensemble models formed by individual technologies. Our goal in this paper is to study the performance of different ensembles of cloud-based technologies for sentiment analysis. Overall, we find that ensemble models perform better on long texts, a scenario where individual technologies tend to struggle. We also find that score-based ensembles perform better than label-based ensembles. Besides being of value to practitioners, we discuss how our results might increase the reliability of research findings that rely on sentiment analysis. In particular, we argue that ensemble models may reduce the chances of sentiment-dependent results being driven by a single technology.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.