Quantifying affective intensity in social media text-based content: The expressive feature fusion framework

Author Identifier (ORCID)

Saiyidi Mat Roni: https://orcid.org/0000-0001-9511-7786

Violetta Wilk: https://orcid.org/0000-0001-7990-769X

Pi Shen Seet: https://orcid.org/0000-0002-0267-5947

Abstract

The increasing prevalence of user generated content (UGC) on social media has intensified the need for sentiment analysis models capable of handling informal, expressive language. Conventional sentiment analysis approaches tend to underrepresent non-lexical features such as emojis, punctuation, and capitalisation that carry affective intensity. This paper proposes the Expressive Feature Fusion (EFF) framework, a protocol that integrates base sentiment models (e.g., TextBlob, VADER, RoBERTa, and Big Bird Flight 2) with rule-based typographic features. The framework normalises heterogeneous sentiment outputs when required, then computes expressive modifiers, and finally combines them through weighted polarity aggregation. We evaluated the EFF framework on synthetic UGC to demonstrate the framework's ability to improve sentiment scores across lexicon- and transformer-based models. Notably, the EFF framework is able to also enhance sensitivity to affective nuances such as sarcasm, irony, and emoji valence. Our proposed framework presents an interpretable approach for quantifying typographic characteristics in informal text, and opens new pathways for research in marketing analytics, social media monitoring, online narrative analysis, and sentiment analysis in general.

Keywords

BERT, Big Bird Flight 2, RoBERTa, sentiment analysis, social media, TextBlob, tweet, UGC, VADER

Document Type

Conference Proceeding

Date of Publication

1-1-2026

Publication Title

2026 International Conference on Artificial Intelligence, Computer, Data Sciences and Applications (ACDSA)

Publisher

IEEE

School

School of Business and Law

Comments

Roni, S. M., Wilk, V., & Seet, P.-S. (2026). Quantifying affective intensity in social media text-based content: The expressive feature fusion framework. In 2026 International Conference on Artificial Intelligence, Computer, Data Sciences and Applications (ACDSA) (pp. 1–6). IEEE. https://doi.org/10.1109/ACDSA67686.2026.11468059

Copyright

subscription content

Share

 
COinS
 

Link to publisher version (DOI)

10.1109/ACDSA67686.2026.11468059