Situation Context in Context-Aware Emotion Recognition

Abstract

Emotion recognition systems often infer affective states from observable signals such as facial expressions, speech, body posture, language, or multimodal behaviour. Yet emotional meaning is rarely determined by these signals alone: the same expression may communicate different emotions depending on the surrounding situation context. In this review, situation context refers to the external, environmental, temporal, social, conversational, or event-related information that shapes how an emotional signal is interpreted. This paper presents a structured literature review (SLR) synthesizing 19 papers across five search strata to examine how situation context is conceptualised and modelled in context-aware emotion recognition research. The synthesis shows that situation context is not one shared computational object. It is operationalised as visual surroundings, conversational history, social interaction, causal event structure, commonsense knowledge, or semantic situation understanding, depending on the task and modelling tradition. Modelling approaches range from implicit use of temporal or multimodal features to explicit representations through context branches, graph structures, cause labels, commonsense knowledge, or prompting. A central limitation of the field is therefore structural fragmentation: situation context is considered important, but it is defined, represented, and evaluated inconsistently across tasks and datasets.