KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Fine-grained emotion analysis
Advanced sentiment analysis approach that identifies specific emotions rather than general polarities, enabling nuanced understanding of human affective states. This technique uses sophisticated models to distinguish between multiple emotional categories with high accuracy.
Multi-label emotional classification
Machine learning method allowing simultaneous assignment of multiple emotional labels to the same text, recognizing the complex and often multiple nature of human expressions. Unlike single-label classification, it captures the coexistence of different emotions in the same discourse.
Plutchik's model
Psychological theory formalized as a computational model representing eight primary emotions (joy, trust, fear, surprise, sadness, disgust, anger, anticipation) with their intensity variations. This model serves as a theoretical foundation for many emotional classification systems in AI.
Valence-activation space
Two-dimensional representation of emotions where valence (positive-negative) and activation (calm-excited) form a continuous space to characterize affective states. This approach allows more granular modeling of emotions than traditional discrete categories.
Contextual emotional embeddings
Vector representations that capture not only the semantic meaning of words but also their variable emotional charge depending on usage context. These embeddings are essential for understanding context-dependent emotional nuances in natural language processing.
Attention neural network for emotions
Deep learning architecture using attention mechanisms to identify text segments most relevant for detecting specific emotions. This approach allows the model to focus on key words carrying emotional weight in long sentences.
Multimodal emotional fusion
Technique integrating multiple data sources (text, audio, video, facial expressions) to improve emotion detection accuracy. This approach exploits the complementarity of different human communication channels for more robust analysis.
Emotional score calibration
Process of adjusting output probabilities from emotional classification models to better reflect the actual likelihood of each emotion. This calibration is essential for critical applications where confidence in predictions is paramount.
Hierarchical Emotion Classification
Approach structuring emotions into a hierarchy (general categories then specific ones) to improve the coherence and interpretability of predictions. This method captures both broad emotional trends and their specific nuances.
Emotional Transfer Learning
Technique reusing knowledge acquired from large general corpora to adapt models to specific emotion detection tasks with little data. This approach is particularly effective for languages or domains with limited emotional corpora.
Emotional Regularization
Constraint method applied during training to ensure the consistency of emotional predictions and avoid aberrant classifications. It imposes structural relationships between different emotional classes based on cognitive psychology.
Implicit Emotion Detection
Advanced technique identifying emotions not explicitly expressed but implied through context, tone, or linguistic choices. This capability distinguishes sophisticated systems from simple classifiers based on explicit keywords.
Emotional Gaussian Mixture Model
Probabilistic approach modeling the distribution of emotions as a combination of multiple Gaussian distributions in the feature space. This method captures the continuous and often fuzzy nature of boundaries between emotional categories.
Macro-Emotional F1-Score Metric
Evaluation metric calculating the average of the individual F1-score of each emotional class without considering their frequency, thus avoiding biases toward majority classes. This metric is particularly relevant for imbalanced emotional datasets.
Analysis of Emotional Co-occurrences
Statistical study of patterns of simultaneous appearance of different emotions in texts to understand their structural relationships. This analysis helps refine models by exploiting natural dependencies between affective states.