Emotion Recognition In Conversations Research Articles

Deep emotion recognition in textual conversations: a survey

Emotion Recognition in Conversations (ERC) is a key step towards successful human–machine interaction. While the field has seen tremendous advancement in the last few years, new applications and implementation scenarios present novel challenges and opportunities. These range from leveraging the conversational context, speaker, and emotion dynamics modelling, to interpreting common sense expressions, informal language, and sarcasm, addressing challenges of real-time ERC, recognizing emotion causes, different taxonomies across datasets, multilingual ERC, and interpretability. This survey starts by introducing ERC, elaborating on the challenges and opportunities of this task. It proceeds with a description of the emotion taxonomies and a variety of ERC benchmark datasets employing such taxonomies. This is followed by descriptions comparing the most prominent works in ERC with explanations of the neural architectures employed. Then, it provides advisable ERC practices towards better frameworks, elaborating on methods to deal with subjectivity in annotations and modelling and methods to deal with the typically unbalanced ERC datasets. Finally, it presents systematic review tables comparing several works regarding the methods used and their performance. Benchmarking these works highlights resorting to pre-trained Transformer Language Models to extract utterance representations, using Gated and Graph Neural Networks to model the interactions between these utterances, and leveraging Generative Large Language Models to tackle ERC within a generative framework. This survey emphasizes the advantage of leveraging techniques to address unbalanced data, the exploration of mixed emotions, and the benefits of incorporating annotation subjectivity in the learning phase.

MLGAT: multi-layer graph attention networks for multimodal emotion recognition in conversations

MLGAT: multi-layer graph attention networks for multimodal emotion recognition in conversations

HiMul-LGG: A hierarchical decision fusion-based local–global graph neural network for multimodal emotion recognition in conversation

Emotion recognition in conversation (ERC) is a vital task that requires deciphering human emotions through analysis of contextual and multimodal information. However, extant research on ERC concentrates predominantly on investigating multimodal fusion while overlooking the model’s constraints in dealing with unimodal representation discrepancy and speaker dependencies. To address the aforementioned problems, this paper proposes a Hierarchical decision fusion-based Local–Global Graph Neural Network for multimodal ERC (HiMul-LGG). HiMul-LGG employs a hierarchical decision fusion strategy to ensure feature alignment across modalities. Moreover, HiMul-LGG also adopts a local–global graph neural network architecture to reinforce inter-modality and intra-modality speaker dependency. Additionally, HiMul-LGG utilizes a cross-modal multi-head attention mechanism to promote interplay between modalities. We evaluate HiMul-LGG on two emotion recognition datasets, IEMOCAP and MELD, where HiMul-LGG outperforms existing methods. The results of the ablation study also imply the effectiveness of the proposed hierarchical decision fusion strategy and local–global structure of Graph construction.

Correlation mining of multimodal features based on higher-order partial least squares for emotion recognition in conversations

In fields requiring an understanding of emotions, such as digital human interaction and public opinion analysis, achieving a dependable and interpretable model for mining correlations among multimodal features remains a primary objective. However, current deep learning methods often lack transparency and suffer from low interpretability. To address these challenges, we propose a novel Correlation Mining method based on Higher-Order Partial Least Squares (HOPLS) for multimodal Emotion Recognition in conversations (CMHER) in this paper. CMHER innovatively combines HOPLS with transformers and Gated Recurrent Units (GRUs) to compute correlation matrices within unimodal data streams and between cross-modal sources. HOPLS projects source data into a latent space to predict target data via correlation matrix computations, eliminating the need for Graphical Processing Unit (GPU) acceleration and making it suitable for experimental and edge systems. The integration of HOPLS with deep neural networks involves preprocessing multimodal features into suitable dimensions and latent representations, followed by HOPLS computing correlation matrices for cross-modal latent vectors and final labels through optimal joint subspace approximation, which aims at the improvements of both interpretability and reliability. Additionally, a generalization error fitting module further refines the predicted correlation matrices to improve predictive capability and overall model performance. Experiments on two public datasets validate the superiority of our proposed method.

Multimodal graph learning with framelet-based stochastic configuration networks for emotion recognition in conversation

The multimodal emotion recognition in conversation (ERC) task presents significant challenges due to the complexity of relationships and the difficulty in achieving semantic fusion across various modalities. Graph learning, recognized for its capability to capture intricate data relations, has been suggested as a solution for ERC. However, existing graph-based ERC models often fail to address the fundamental limitations of graph learning, such as assuming pairwise interactions and neglecting high-frequency signals in semantically-poor modalities, which leads to an over-reliance on text. While these issues might be negligible in other applications, they are crucial for the success of ERC. In this paper, we propose a novel framework for ERC, namely multimodal graph learning with framelet-based stochastic configuration networks (i.e., Frame-SCN). Specifically, framelet-based stochastic configuration networks, which employ 2D directional Haar framelets to extract both low- and high-pass components, are introduced to learn the unified semantic embeddings from multimodal data, mitigating prediction biases caused by an excessive reliance on text without introducing an unnecessarily large number of parameters. Also, we develop a modality-aware information extraction module that is able to extract both general and sensitive information in a multimodal semantic space, alleviating potential noise issues. Extensive experiment results demonstrate that our proposed Frame-SCN outperforms many state-of-the-art approaches on two widely used multimodal ERC datasets.

Adversarial alignment and graph fusion via information bottleneck for multimodal emotion recognition in conversations

With the rapid development of social media and human–computer interaction, multimodal emotion recognition in conversations (MERC) tasks have begun to receive widespread research attention. The MERC task is to extract and fuse complementary semantic information from different modalities to classify the speaker’s emotion. However, the existing feature fusion methods usually directly map the features of other modalities into the same feature space for information fusion, which cannot eliminate the heterogeneity between different modalities and make the subsequent emotion class boundary learning more difficult. In addition, existing graph contrastive learning methods obtain consistent feature representations by maximizing mutual information between multiple views, which may lead to overfitting of the model. To tackle the above problem, we propose a novel Adversarial Alignment and Graph Fusion via Information Bottleneck for Multimodal Emotion Recognition in Conversations (AGF-IB) method. Firstly, we input video, audio, and text features into a multi-layer perceptron (MLP) to map them into separate feature spaces. Secondly, we build a generator and a discriminator for the three modal features, respectively, through adversarial representation to achieve information interaction between modalities and eliminate the heterogeneity among modalities. Thirdly, we introduce graph contrastive representation learning to capture intra-modal and inter-modal complementary semantic information and learn intra-class and inter-class boundary information of emotion categories. Furthermore, instead of maximizing the mutual information (MI) between multiple views, we use information bottleneck theory to minimize the MI between views. Specifically, we construct a graph structure for the three modal features respectively and perform contrastive representation learning on nodes with different emotions in the same modality and nodes with the same emotion in different modalities, to improve the feature representation ability of nodes. Finally, we use MLP to complete the emotional classification of the speaker. Extensive experiments show that AGF-IB can improve emotion recognition accuracy on IEMOCAP and MELD datasets. Furthermore, since AGF-IB is a general multimodal fusion and contrastive learning method, it can be applied to other multimodal tasks in a plug-and-play manner, e.g., humor detection.

Multi-modal graph context extraction and consensus-aware learning for emotion recognition in conversation

Multi-modal emotion recognition in conversation is challenging because of the difficulty to jointly leverage the information from heterogeneous text, acoustic, and visual modalities. Recent context-aware methods usually design a graph structure to model dependencies of utterances and speakers, or integrate Multi-modal information. However, they typically lack a sufficient extraction of unimodal context, and rarely explore the emotion consensus prototypes among different samples with the same label. For solving these problems, in this paper, we propose a Graph Context extraction and Consensus-aware Learning (GCCL) framework to excavate context-sensitive fusion features and simulate the emotion evocation process during the emotion consensus learning. Specifically, GCCL contains a well-designed graph-based module to capture speaker, temporal and modality dependencies and integrate information from different modalities. Then, we design an emotion consensus learning unit to mine the most typical feature of each category in each modality. A speaker-guided contrastive learning loss is further proposed to guarantee the diversity between different individuals and the semantic consistency between distinct modalities. Moreover, we construct a consensus-aware unit with an attention-based memory mechanism to preserve semantic correlations among different samples on the category-level. Extensive experimental results on two conversational datasets demonstrate that the proposed GCCL outperforms the state-of-art methods. Code is available at https://github.com/gityider/GCCL.

Speaker-aware cognitive network with cross-modal attention for multimodal emotion recognition in conversation

Emotion recognition in conversation (ERC) has gained considerable attention owing to its extensive applications in the field of human-computer interaction. However, previous models have had certain limitations in exploring the potential emotional relationships within the conversation due to their inability to fully leverage speaker information. Additionally, information from various modalities such as text, audio, and video can synergistically enhance and supplement the analysis of emotional context within the conversation. Nonetheless, effectively fusing multimodal features to understand the detailed contextual information in the conversation is challenging. This paper proposes a Speaker-Aware Cognitive network with Cross-Modal Attention (SACCMA) for multimodal ERC to effectively leverage multimodal information and speaker information. Our proposed model primarily consists of the modality encoder and the cognitive module. The modality encoder is employed to fuse multimodal feature information from speech, text, and vision using a cross-modal attention mechanism. Subsequently, the fused features and speaker information are separately fed into the cognitive module to enhance the perception of emotions within the dialogue. Compared to seven common baseline methods, our model increased the Accuracy score by 2.71 % and 1.70 % on the IEMOCAP and MELD datasets, respectively. Additionally, the F1 score improved by 2.92 % and 0.70 % for each dataset. Various experiments also demonstrate the effectiveness of our method.

Disentangled Variational Autoencoder for Emotion Recognition in Conversations

In Emotion Recognition in Conversations (ERC), the emotions of target utterances are closely dependent on their context. Therefore, existing works train the model to generate the response of the target utterance, which aims to recognise emotions leveraging contextual information. However, adjacent response generation ignores long-range dependencies and provides limited affective information in many cases. In addition, most ERC models learn a unified distributed representation for each utterance, which lacks interpretability and robustness. To address these issues, we propose a VAD-disentangled Variational AutoEncoder (VAD-VAE), which first introduces a target utterance reconstruction task based on Variational Autoencoder, then disentangles three affect representations Valence-Arousal-Dominance (VAD) from the latent space. We also enhance the disentangled representations by introducing VAD supervision signals from a sentiment lexicon and minimising the mutual information between VAD distributions. Experiments show that VAD-VAE outperforms the state-of-the-art model on two datasets. Further analysis proves the effectiveness of each proposed module and the quality of disentangled VAD representations. The code is available at https://github.com/SteveKGYang/VAD-VAE.

DC-BVM: Dual-channel information fusion network based on voting mechanism

Emotion recognition in conversations (ERC) has been challenging due to the dynamics and complexity of emotions in conversations. Most current emotion recognition studies have focused on modeling temporal dimensions, such as context-sensitive dependencies while ignoring the spatial dimensional relationships between discourses. In this paper, we propose a dual-channel information fusion network based on a voting mechanism, in which we model modal information in two dimensions: distance dependence is constructed in the spatial dimension, and context dependence is constructed in the temporal dimension. We aim to extract more comprehensive and accurate sentiment features to recognize verbal emotions from limited data. In addition, we discard the traditional modal fusion method and propose a fusion strategy based on the voting mechanism, which significantly accelerates the model convergence speed and recognition performance. We conducted experiments on two benchmark datasets, IEMOCAP and MELD, respectively, and the experimental results demonstrate the superiority of the designed network in emotion recognition.

Adaptive Graph Learning for Multimodal Conversational Emotion Detection

Multimodal Emotion Recognition in Conversations (ERC) aims to identify the emotions conveyed by each utterance in a conversational video. Current efforts encounter challenges in balancing intra- and inter-speaker context dependencies when tackling intra-modal interactions. This balance is vital as it encompasses modeling self-dependency (emotional inertia) where speakers' own emotions affect them and modeling interpersonal dependencies (empathy) where counterparts' emotions influence a speaker. Furthermore, challenges arise in addressing cross-modal interactions that involve content with conflicting emotions across different modalities. To address this issue, we introduce an adaptive interactive graph network (IGN) called AdaIGN that employs the Gumbel Softmax trick to adaptively select nodes and edges, enhancing intra- and cross-modal interactions. Unlike undirected graphs, we use a directed IGN to prevent future utterances from impacting the current one. Next, we propose Node- and Edge-level Selection Policies (NESP) to guide node and edge selection, along with a Graph-Level Selection Policy (GSP) to integrate the utterance representation from original IGN and NESP-enhanced IGN. Moreover, we design a task-specific loss function that prioritizes text modality and intra-speaker context selection. To reduce computational complexity, we use pre-defined pseudo labels through self-supervised methods to mask unnecessary utterance nodes for selection. Experimental results show that AdaIGN outperforms state-of-the-art methods on two popular datasets. Our code will be available at https://github.com/TuGengs/AdaIGN.

Multimodal Knowledge-enhanced Interactive Network with Mixed Contrastive Learning for Emotion Recognition in Conversation

Emotion Recognition in Conversations (ERC) aims to accurately identify the emotional labels of each utterance in a conversation, holding significant application value in human–computer interaction. Existing research suggests introducing commonsense knowledge (CSK) and multimodal information enhances model performance in ERC tasks. However, several challenges persist: (1) the neglect of complex psychological influences between utterances; (2) noise issues within modal information; (3) prediction challenges for emotion labels with few samples in different categories that exhibit semantic similarity but distinct emotional categories. To address the above problems, we propose a Multimodal Knowledge-enhanced Interactive Network with Mixed Contrastive Learning (MKIN-MCL). Firstly, we establish a knowledge aggregation graph to capture the dependencies of commonsense knowledge (CSK) between utterances during a conversation. We actively aggregate relevant knowledge information to enhance text features. Simultaneously, we apply feature filters for acoustic and visual modalities to eliminate noise and enhance feature quality. Furthermore, we implement an interactive attention module by stacking designed Cross-modal Interactive Transformers (CITs) to continuously explore the relevance between the interacting parties in their respective semantic spaces, thus improving the effectiveness of modality interaction while reducing noise generated during the interaction. Lastly, we employ the Mixed Contrastive Learning (MCL) strategy to enhance the model’s ability to handle few-shot labels. This strategy utilizes unsupervised contrastive learning to improve the representation capability of the multimodal fusion features and supervised contrastive learning to extract information from few-shot labels. Extensive experiments on two benchmark datasets, IEMOCAP and MELD, validate the effectiveness and superiority of the proposed model.

MRSLN: A Multimodal Residual Speaker-LSTM Network to alleviate the over-smoothing issue for Emotion Recognition in Conversation

Multimodal Emotion Recognition in Conversation (ERC) plays a significant role in the field of human–computer intelligent interaction since it enables computers to perceive and infer the emotions expressed by the individuals, thereby intelligently responding to them. Most of current ERC methods pay more attention to modeling the complex interaction between different modalities. However, the features extracted by their unimodal networks are over-smoothed and may contain insufficient intra-speaker contextual information, which results in suboptimal results. In this paper, we focus on the unimodal learning and propose a simple late fusion framework named Multimodal Residual Speaker-LSTM Network (MRSLN), which uses speaker information to directly model inter-speaker and intra-speaker dependency, rather than fuse it into the learned features. MRSLN uses the speaker-LSTM consisting of the inter-speaker LSTM, intra-speaker LSTM, and the residual network between the input and output of the inter-speaker LSTM. Our proposed method can alleviate the issue of over-smoothing in deep Long Short Term Memory (LSTM) network and also incorporate additional intra-speaker contextual information. Extensive experiments conducted on IEMOCAP and MELD datasets demonstrate that MRSLN effectively captures inter-speaker and intra-speaker information and outperforms currently complex state-of-the-art (SOTA) models in efficiency and classification performance.

Multi-Modal Attentive Prompt Learning for Few-shot Emotion Recognition in Conversations

Emotion recognition in conversations (ERC) has emerged as an important research area in Natural Language Processing and Affective Computing, focusing on accurately identifying emotions within the conversational utterance. Conventional approaches typically rely on labeled training samples for fine-tuning pre-trained language models (PLMs) to enhance classification performance. However, the limited availability of labeled data in real-world scenarios poses a significant challenge, potentially resulting in diminished model performance. In response to this challenge, we present the Multi-modal Attentive Prompt (MAP) learning framework, tailored specifically for few-shot emotion recognition in conversations. The MAP framework consists of four integral modules: multi-modal feature extraction for the sequential embedding of text, visual, and acoustic inputs; a multi-modal prompt generation module that creates six manually-designed multi-modal prompts; an attention mechanism for prompt aggregation; and an emotion inference module for emotion prediction. To evaluate our proposed model’s efficacy, we conducted extensive experiments on two widely recognized benchmark datasets, MELD and IEMOCAP. Our results demonstrate that the MAP framework outperforms state-of-the-art ERC models, yielding notable improvements of 3.5% and 0.4% in micro F1 scores. These findings highlight the MAP learning framework’s ability to effectively address the challenge of limited labeled data in emotion recognition, offering a promising strategy for improving ERC model performance.

Hypergraph Neural Network for Emotion Recognition in Conversations

Modeling conversational context is an essential step for emotion recognition in conversations. Existing works still suffer from insufficient utilization of local context information and remote context information. This article designs a hypergraph neural network, namely HNN-ERC, to better utilize local and remote contextual information. HNN-ERC combines the recurrent neural network with the conventional hypergraph neural network to strengthen connections between utterances and make each utterance receive information from other utterances better. The proposed model has empirically achieved state-of-the-art results on three benchmark datasets, demonstrating the effectiveness and superiority of the new model.

Self-supervised utterance order prediction for emotion recognition in conversations

As the order of the utterances in a conversation changes, the meaning of the utterance also changes, and sometimes, this will cause different semantics or emotions. However, the existing representation learning models do not pay close attention to capturing the internal semantic differences of utterance caused by the change of utterance order. Based on this, we build a self-supervised utterance order prediction approach to learn the logical order of utterance, which helps understand the deep semantic relationship between adjacent utterances. Specially, the utterance binary composed of two adjacent utterances, which are ordered or disordered, is fed to the self-supervised model so that the self-supervised model can obtain firm representation learning ability for the semantic differences of the adjacent sentences. The self-supervised method is applied to the downstream conversation emotion recognition task to test the value of the approach. The features extracted from the self-supervised model are fused with the multimodal features to obtain a richer utterance representation. After that, emotion recognition models are applied to two different datasets. The experiment results show that our proposed approach outperforms the current state of the art on ERC benchmark datasets.

PIRNet: Personality-Enhanced Iterative Refinement Network for Emotion Recognition in Conversation.

Emotion recognition in conversation (ERC) is important for enhancing user experience in human-computer interaction. Unlike vanilla emotion recognition in individual utterances, ERC aims to classify constituent utterances in a dialog into corresponding emotion labels, which makes contextual information crucial. In addition to contextual information, personality traits also affect emotional perception based on psychological findings. Although researchers have proposed several approaches and achieved promising results on ERC, current works in this domain rarely incorporate contextual information and personality influence. To this end, we propose a novel framework to integrate these factors seamlessly, called "Personality-enhanced Iterative Refinement Network (PIRNet)." Specifically, PIRNet is a multistage iterative method. To capture personality influence, PIRNet leverages personality traits to mimic emotional transitions and generates personality-enhanced results. Then we exploit sequence models to capture contextual information in conversations. To verify the effectiveness of our proposed method, we conduct experiments on three benchmark datasets for ERC, that is, IEMOCAP, CMU-MOSI, and CMU-MOSEI. Experimental results demonstrate that our PIRNet succeeds over currently advanced approaches to emotion recognition.

ERNetCL: A novel emotion recognition network in textual conversation based on curriculum learning strategy

Emotion recognition in conversation (ERC) has emerged as a research hotspot in domains such as conversational robots and question-answer systems. How to efficiently and adequately retrieve contextual emotional cues has been one of the key challenges in the ERC task. Existing efforts do not fully model the context and employ complex network structures, resulting in limited performance gains. In this paper, we propose a novel emotion recognition network based on curriculum learning strategy (ERNetCL). The proposed ERNetCL primarily consists of temporal encoder (TE), spatial encoder (SE), and curriculum learning (CL) loss. We utilize TE and SE to combine the strengths of previous methods in a simplistic manner to efficiently capture temporal and spatial contextual information in the conversation. To ease the harmful influence resulting from emotion shift and simulate the way humans learn curriculum from easy to hard, we apply the idea of CL to the ERC task to progressively optimize the network parameters. At the beginning of training, we assign lower learning weights to difficult samples. As the epoch increases, the learning weights for these samples are gradually raised. Extensive experiments on four datasets exhibit that our proposed method is effective and dramatically beats other baseline models.

Modeling Hierarchical Uncertainty for Multimodal Emotion Recognition in Conversation.

Approximating the uncertainty of an emotional AI agent is crucial for improving the reliability of such agents and facilitating human-in-the-loop solutions, especially in critical scenarios. However, none of the existing systems for emotion recognition in conversation (ERC) has attempted to estimate the uncertainty of their predictions. In this article, we present HU-Dialogue, which models hierarchical uncertainty for the ERC task. We perturb contextual attention weight values with source-adaptive noises within each modality, as a regularization scheme to model context-level uncertainty and adapt the Bayesian deep learning method to the capsule-based prediction layer to model modality-level uncertainty. Furthermore, a weight-sharing triplet structure with conditional layer normalization is introduced to detect both invariance and equivariance among modalities for ERC. We provide a detailed empirical analysis for extensive experiments, which shows that our model outperforms previous state-of-the-art methods on three popular multimodal ERC datasets.

GraphCFC: A Directed Graph Based Cross-Modal Feature Complementation Approach for Multimodal Conversational Emotion Recognition

Emotion Recognition in Conversation (ERC) plays a significant part in Human-Computer Interaction (HCI) systems since it can provide empathetic services. Multimodal ERC can mitigate the drawbacks of uni-modal approaches. Recently, Graph Neural Networks (GNNs) have been widely used in a variety of fields due to their superior performance in relation modeling. In multimodal ERC, GNNs are capable of extracting both long-distance contextual information and inter-modal interactive information. Unfortunately, since existing methods such as MMGCN directly fuse multiple modalities, redundant information may be generated and diverse information may be lost. In this work, we present a directed Graph based Cross-modal Feature Complementation (GraphCFC) module that can efficiently model contextual and interactive information. GraphCFC alleviates the problem of heterogeneity gap in multimodal fusion by utilizing multiple subspace extractors and Pair-wise Cross-modal Complementary (PairCC) strategy. We extract various types of edges from the constructed graph for encoding, thus enabling GNNs to extract crucial contextual and interactive information more accurately when performing message passing. Furthermore, we design a GNN structure called GAT-MLP, which can provide a new unified network framework for multimodal learning. The experimental results on two benchmark datasets show that our GraphCFC outperforms the state-of-the-art (SOTA) approaches.