Directed Acyclic Graph Network for Conversational Emotion Recognition

Shen, Weizhou; Wu, Siyue; Yang, Yunyi

doi:10.48550/arxiv.2105.12907

Cited by 12 publications

(25 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…GNN-based methods: DialogurGCN (Ghosal et al 2019), RGAT (Ishiwatari et al 2020) and DAG-ERC (Shen et al 2021).…”

Section: Baseline Methodsmentioning

confidence: 99%

“…In DialogueGCN (Ghosal et al 2019) and RGAT (Ishiwatari et al 2020), GCN (Kipf and Welling 2016) and GAT (Veličković et al 2017) are applied to refine the features with speaker dependencies and temporal information. DAG-ERC (Shen et al 2021) applies a directed acyclic graph for conversation representation and it achieves the state-of-the-art performance on multiple ERC datasets. Vaswani et al (2017) first propose Transformer for machine translation task, whose success subsequently has been proved in various down-stream NLP tasks.…”

Section: Emotion Recognition In Conversationmentioning

confidence: 99%

“…For the evaluation metrics, we choose micro-averaged F1 for DailyDialog and weighted-average F1 for the other datasets, following previous works (Ishiwatari et al 2020;Shen et al 2021;Zhong, Wang, and Miao 2019).…”

Section: Datasetsmentioning

confidence: 99%

“…Instead of modeling only sequential data recurrently in RNN, GNN is designed to capture all kinds of graph structure information via various aggregation algorithms. Existing GNN-based ERC methods, such as Di-alogueGCN (Ghosal et al 2019), RGAT (Ishiwatari et al 2020) and DAG-ERC (Shen et al 2021), which are the state of the art, have demonstrated the superiority of GNN in modeling conversational context. A directed graph is constructed on each dialogue in these methods, where the nodes denote the individual utterances, and the edges indicate relationships between utterances.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

S+PAGE: A Speaker and Position-Aware Graph Neural Network Model for Emotion Recognition in Conversation

Chen¹,

Chong²,

Xu³

et al. 2021

Preprint

View full text Add to dashboard Cite

Emotion recognition in conversation (ERC) has attracted much attention in recent years for its necessity in widespread applications. Existing ERC methods mostly model the self and inter-speaker context separately, posing a major issue for lacking enough interaction between them. In this paper, we propose a novel Speaker and Position-Aware Graph neural network model for ERC (S+PAGE), which contains three stages to combine the benefits of both Transformer and relational graph convolution network (R-GCN) for better contextual modeling. Firstly, a two-stream conversational Transformer is presented to extract the coarse self and inter-speaker contextual features for each utterance. Then, a speaker and position-aware conversation graph is constructed, and we propose an enhanced R-GCN model, called PAG, to refine the coarse features guided by a relative positional encoding. Finally, both of the features from the former two stages are input into a conditional random field layer to model the emotion transfer. Extensive experiments demonstrate that our model achieves state-of-the-art performance on three ERC datasets.

show abstract

“…GNN-based methods: DialogurGCN (Ghosal et al 2019), RGAT (Ishiwatari et al 2020) and DAG-ERC (Shen et al 2021).…”

Section: Baseline Methodsmentioning

confidence: 99%

Section: Emotion Recognition In Conversationmentioning

confidence: 99%

Section: Datasetsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

S+PAGE: A Speaker and Position-Aware Graph Neural Network Model for Emotion Recognition in Conversation

Chen¹,

Chong²,

Xu³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Proposed by Ghosal et al [10], COSMIC modeled interactions between the interlocutors within a conversation based on different elements of commonsense. Recently, Shen et al [27] combined the advantages of graph-based neural models and recurrence-based neural models to design a directed acyclic neural network to model the intrinsic structure of dialogue.…”

Section: Sentiment Analysismentioning

confidence: 99%

Multimodal Representations Learning Based on Mutual Information Maximization and Minimization and Identity Embedding for Multimodal Sentiment Analysis

Zheng¹,

Zhang²,

Wang³

et al. 2022

Preprint

View full text Add to dashboard Cite

Multimodal sentiment analysis (MSA) is a fundamental complex research problem due to the heterogeneity gap between different modalities and the ambiguity of human emotional expression. Although there have been many successful attempts to construct multimodal representations for MSA, there are still two challenges to be addressed: 1) A more robust multimodal representation needs to be constructed to bridge the heterogeneity gap and cope with the complex multimodal interactions, and 2) the contextual dynamics must be modeled effectively throughout the information flow. In this work, we propose a multimodal representation model based on Mutual information Maximization and Minimization and Identity Embedding (MMMIE). We combine mutual information maximization between modal pairs, and mutual information minimization between input data and corresponding features to mine the modal-invariant and task-related information. Furthermore, Identity Embedding is proposed to prompt the downstream network to perceive the contextual information. Experimental results on two public datasets demonstrate the effectiveness of the proposed model.

show abstract

DialoguePCN: Perception and Cognition Network for Emotion Recognition in Conversations

Wu,

Feng,

et al. 2023

IEEE Access

View full text Add to dashboard Cite

In the Emotion Recognition in Conversations (ERC) task, extracting emotional cues from the context is an effective strategy for improving model performance. However, current research has two evident limitations: firstly, irrelevant context information severely affects the extraction of emotional features at the utterance level. Secondly, in dialogues, subsequent utterances' retrieval of emotional cues does not benefit from extracted emotional cues from preceding utterances. This paper designs a Dialogue Perception Cognition Network (DialoguePCN) model, which aims to solve the issues above by simulating the perception and cognition phases of emotion in conversations. In the perception phase, DialoguePCN proposes an activation module based on a cosine similarity selection algorithm, providing a dynamic initial emotional state for the predicted utterance. In the cognition phase, the model introduces a new gating mechanism, marking the first attempt to use the extracted utterance emotion representation to reconstruct context information iteratively. This approach reduces the complexity of retrieving emotional cues from the context and solves the inherent cold-start challenge in ERC tasks. Using audio and text features, the accuracy of DialoguePCN reached 68.7% on the IEMOCAP dataset.

show abstract

Directed Acyclic Graph Network for Conversational Emotion Recognition

Cited by 12 publications

References 24 publications

S+PAGE: A Speaker and Position-Aware Graph Neural Network Model for Emotion Recognition in Conversation

S+PAGE: A Speaker and Position-Aware Graph Neural Network Model for Emotion Recognition in Conversation

Multimodal Representations Learning Based on Mutual Information Maximization and Minimization and Identity Embedding for Multimodal Sentiment Analysis

DialoguePCN: Perception and Cognition Network for Emotion Recognition in Conversations

Contact Info

Product

Resources

About