Emotion recognition in conversation (ERC) is an important research direction in the field of human-computer interaction (HCI), which recognizes emotions by analyzing utterance signals to enhance user experience and plays an important role in several domains. However, existing research on ERC mainly focuses on constructing graph networks by directly modeling interactions on multimodal fused features, which cannot adequately capture the complex dialog dependency based on time, speaker, modalities, etc. In addition, existing multi-task learning frameworks for ERC do not systematically investigate how and where gender information is injected into the model to optimize ERC performance. To address the above problems, this paper proposes a Hierarchical Graph Fusion for ERC with Mid-Late Gender-aware Strategy (HGF-MiLaG). HGF-MiLaG uses hierarchical fusion graph to adequately capture intra-modal and inter-modal speaker dependency and temporal dependency. In addition, HGF-MiLaG explores the effect of the location of gender information injections on ERC performance, and ultimately employs a Mid-Late multilevel gender-aware strategy in order to allow the hierarchical graph network to determine the proportion of emotion and gender information in the classifier. Empirical results on two public multimodal datasets(i.e.,IEMOCAP and MELD), demonstrate that HGF-MiLaG outperforms existing methods.
Speech emotion recognition (SER) infers affective states from speech signals, but positional encoding for acoustic tokens remains underexplored in Transformer-based SER. Existing models often reuse encodings designed for text and do not explicitly account for the different sequential and two-dimensional structures of acoustic representations. We propose Fibonacci Position Embedding (FPE) and Fibonacci Target Shutter (FTS). FTS constructs overlapping candidate-index sets over time–frequency token grids, and FPE samples a Fibonacci index and applies its modulo-wrapped, dimension-dependent phase rotation to query and key vectors. The modules are integrated into STRE-Former, which fuses Wav2Vec, log-mel spectrogram, and MFCC representations through asymmetric cross-representation attention with representation-specific positional encodings. We also introduce an implementation-consistent conditional-entropy formulation that quantifies uncertainty in recovering a token location from its sampled positional representation; this quantity characterizes positional ambiguity rather than downstream modeling capacity. Experiments over 64 positional-encoding combinations on IEMOCAP and MELD identify dataset-dependent highest-observed configurations, reaching 74.21% weighted accuracy on IEMOCAP-4, 74.54% on IEMOCAP-6, and 49.44% on MELD. These empirical observations suggest that the relative behavior of positional-encoding strategies may depend on the acoustic representation and evaluation dataset, rather than supporting a single universally optimal scheme.