KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Sentiment Inference
Process of deducing an emotion or opinion from contextual and non-explicit clues when polarity words are absent from the text.
Implicit Polarity
Positive or negative orientation of a text that is not directly expressed by adjectives or emotional keywords, but inferred from the described situation.
Contextual Inference Model
Machine learning architecture, often based on transformers, designed to understand long-term relationships and global context to reveal hidden sentiments.
Connotation Analysis
Examination of word associations and secondary meanings of neutral words to reveal emotional charge or indirect opinion in discourse.
Abductive Reasoning
Logical inference method for determining the best possible explanation for a set of observations, used to guess the most likely sentiment of an ambiguous text.
Computational Sarcasm
Branch of natural language processing that develops algorithms to detect sarcasm, a form of implicitness where positive statements mask negative criticism.
World Knowledge Model
System integrating factual databases and commonsense schemas to enable AI to make logical deductions about indirectly described emotional states.
Inter-sentence Analysis
Technique that goes beyond isolated sentences to analyze relationships and contrasts between multiple sentences to infer overall opinion or sentiment changes.
Detection of veiled opinion
Identification of opinions expressed through facts, neutral descriptions, or rhetorical questions, where the value judgment is implied rather than stated.
Entity-based inference
Method that deduces sentiment by analyzing the relationships between named entities (people, products, brands) and the actions or states associated with them in the text.
Reasoning graph model
Approach that structures text information in the form of a graph (nodes and edges) to model the logical inferences necessary for detecting implicit sentiments.
Modality analysis
Examination of markers of certainty, possibility, or obligation (e.g., 'could', 'seems', 'must') to assess the speaker's degree of commitment and nuance the inferred sentiment.
Prompt engineering for implicit content
Design of sophisticated instructions for large language models, guiding them to perform multi-step reasoning to detect non-explicit opinions.
Implicit quadrant
Classification framework for texts with implicit sentiment along two axes: the level of ambiguity (low/high) and the source of inference (local context/world knowledge).
Metaphor analysis
Process of interpreting metaphorical expressions to extract their emotional charge or value judgment, which are often masked by figurative meaning.
Euphemism detection
Identification of the use of softened or indirect expressions to replace a potentially negative or unpleasant term, in order to reveal the concealed real sentiment.