A comparative analysis of detection mechanisms for emotion detection

***

Abstract:

This paper compared the performance of emotion detection mechanisms using dataset crawled from Facebook diabetes support group pages. 

To be specific, string-based Multinomial Naïve Bayes algorithm, NRC Emotion Lexicon (Emolex) and Indico API were used to detect five emotions present in 2475 Facebook posts, namely, fear, joy, sad, anger and surprise. 

Both accuracy and F-score measures were used to assess the effectiveness of the algorithms in detecting the emotions. 

Findings indicate string-based Multinomial Naïve Bayes to outperform both Emolex (i.e. 82% vs. 78%) and Indico API (i.e. 82% vs. 50%). 

Further analysis also revealed emotions such as joy, fear and sadness to be of the highest frequencies for the diabetes community. 

Implications of the findings and emotions detected are further discussed in this paper.

*** 

https://iopscience.iop.org/article/10.1088/1742-6596/1339/1/012016

https://iopscience.iop.org/article/10.1088/1742-6596/1339/1/012016/pdf