JMIR Publications

ABSTRACT

Background:

Natural language processing (NLP) and artificial intelligence (AI) models are increasingly used to extract emotional signals from textual data; however, systematic validation of different computational paradigms in emotionally complex relational contexts remains limited.

Objective:

This study aimed to compare a lexicon-based approach (NRC Emotion Lexicon) and a transformer-based model (Pysentimiento) in detecting affective patterns within spontaneous romantic relationship narratives and to evaluate their external validity against a theory-driven measure of relational well-being.

Methods:

A total of 112 Spanish adults completed the Couple Flourishing Measure (CFM) and provided an open-ended description of their romantic relationship. Narratives were analyzed using both computational approaches to generate emotion and sentiment indices. Pearson correlations were computed between model outputs and CFM scores. Steiger’s Z tests were conducted to compare dependent correlations across models.

Results:

Both approaches showed significant associations with relational well-being; however, the transformer-based model demonstrated stronger and more theoretically coherent correlations, particularly for positive and negative valence (r = .66 and r = –.74, respectively). The lexicon-based approach detected smaller associations, primarily within negative emotion dimensions. Significant differences between paradigms were observed across multiple affective categories.

Conclusions:

Transformer-based NLP models show greater sensitivity to emotionally nuanced relational discourse compared to traditional lexicon-based methods. These findings extend methodological evaluation frameworks for computational emotion detection by incorporating theory-driven external validation and highlight implications for responsible AI deployment in psychologically sensitive domains.