From:  Artificial intelligence strategies for emotion recognition in cancer pain research

 NLP approaches for affective text analysis.

Approach/ModelDescriptionStrengths (relevance to APA)Main limitationsRef.
Transformer-based models (BERT, RoBERTa)Pre-trained language models fine-tuned on emotion-specific datasets to capture contextual semantic representationsHigh sensitivity to subtle affective cues; effective modeling of context-dependent emotional languageLimited interpretability; sensitive to domain shift and individual linguistic variability[52, 54]
Lexicon-based and hybrid approaches (e.g., NRC, emoji lexicons)Combination of emotion lexicons and data-driven representations to infer affective content from textInterpretable and robust across domains; useful for capturing multidimensional emotional signalsLimited ability to model complex, implicit, or metaphorical language[55, 56]
Idiographic versus nomothetic modeling approachesComparison between population-level models and personalized language-emotion mappingsEnables modeling of individual variability in emotional expression; critical for personalized APARequires longitudinal and subject-specific data; limited generalizability[53]
Clinical NLP and longitudinal monitoring approachesAnalysis of patient-generated text (e.g., PROs, clinical notes, EMA) to track emotional states over timeSupports early detection of emotional distress; enables continuous and real-world monitoringEthical concerns (privacy); variability in data quality; challenges in temporal consistency[53, 58, 59]
Return to article view

APA: automatic pain assessment; NLP: natural language processing; BERT: Bidirectional Encoder Representations from Transformers; RoBERTa: Robustly Optimized BERT Pretraining Approach; NRC: National Research Council (Emotion Lexicon); EMA: Ecological Momentary Assessment; PROs: patient-reported outcomes.