Researchers have developed a new method called DiaRelay for emotion recognition in conversations. The approach uses a constant-size memory to store dialogue context and allows earlier emotional cues to influence later predictions without re-encoding the entire conversation history. This is achieved through two extra components: Selective Relay Memory Transition, which aggregates historical evidence into a bounded relay memory, and Dual-axis Relay Memory Read, which dynamically modulates feature transformations based on the propagated memory. The method has been tested on two datasets and shows state-of-the-art performance with only a small increase in trainable parameters. --- Why it matters: This work matters to AI researchers because it addresses a common limitation of existing methods for emotion recognition in conversations, where models often discard or re-encode overlapping context. DiaRelay provides a more efficient way to incorporate dialogue history and adapt to changing conversational contexts, which can improve the accuracy of emotional understanding in various applications. Source: https://arxiv.org/abs/2608.22745