Embedding Text in Emotion and Evocation Vector Spaces

ACL ARR 2025 February Submission3959 Authors

Abstract: The aim of this paper is to present the methodology for embedding text in a concrete emotion vector space, where each dimension represents a single emotion, and the coordinates represent emotion intensities. Additionally, the text is also embedded in an experimental (emotion) evocation vector space, where the coordinates represent possible evocation intensities of each emotion for the text reader. Embeddings are performed using newly prepared sentence transformer-based language models, trained on an existing dataset of social media posts written in Serbian and manually labeled with emotions.

Paper Type: Short

Research Area: Language Modeling

Research Area Keywords: Computational Social Science and Cultural Analytics,Efficient/Low-Resource Methods for NLP,Language Modeling

Contribution Types: NLP engineering experiment, Approaches to low-resource settings

Languages Studied: Serbian,Language independant

Submission Number: 3959