Implementing Dialogue Emotion Recognition via Recursive Propagation of Explicit Emotions

In the field of Emotion Recognition in Conversation (ERC), the primary objective is to discern the emotion expressed in each utterance within a dialogue. Traditional methods predominantly analyze the semantic content of conversations, often overlooking the influence of surrounding utterances’ emotions on a given utterance. This oversight may compromise the coherence of sentiment analysis, diminishing both its accuracy and reliability. To counteract these limitations, this study introduces a novel methodology named implementing dialogue emotion recognition via recursive propagation of explicit emotions (RPEE). RPEE integrates a emotional contagion module with a emotional inheritance module, thereby remedying the deficiencies of conventional approaches. The emotional contagion module focuses on identifying and effectively disseminating prominent emotions throughout the conversation, thereby enhancing context awareness in sentiment analysis. Whereas, the emotional inheritance module concentrates on capturing the emotional continuity and progression across consecutive utterances by the same speaker. Both modules utilize advanced emotion pseudo-label embedding techniques to better comprehend intricate emotion dynamics. The distinctiveness of RPEE lies in its simultaneous consideration of both the semantic content of utterances and the dynamic interplay of emotional shifts within dialogues. Rigorous testing on four public datasets substantiates the method's efficacy in elevating the precision of dialog sentiment analysis. Our code is available at https://anonymous.4open.science/r/RPEE-BCEB