Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AIMA at SemEval-2024 Task 3: Simple Yet Powerful Emotion Cause Pair Analysis

Alireza Ghahramani Kure, Mahshid Dehghani, Mohammad Mahdi Abootorabi, Nona Ghazizadeh, Seyed Arshan Dalili, Ehsaneddin Asgari

TL;DR

This work tackles emotion-cause pair extraction in conversations through a text-centric three-component architecture: embedding extraction with EmoBERTa, cause-pair extraction plus emotion classification via a Transformer encoder that yields a causality matrix, and QA-driven extraction of cause spans after pair detection. Although designed to handle multimodal data, the study focuses on textual data for Subtask 1 and demonstrates competitive performance in SemEval-2024 Task 3, ranking 10th in Subtask 1 and 5th in Subtask 2 among 23 teams. Evaluations rely on F1-based metrics, including strict and proportional matching for textual spans, and emphasize the trade-off between emotion recognition accuracy and precise cause extraction. The results suggest that simple, efficient modeling can yield strong insights into conversational emotions while pointing to avenues for lightweight multimodal enhancements in future work.

Abstract

The SemEval-2024 Task 3 presents two subtasks focusing on emotion-cause pair extraction within conversational contexts. Subtask 1 revolves around the extraction of textual emotion-cause pairs, where causes are defined and annotated as textual spans within the conversation. Conversely, Subtask 2 extends the analysis to encompass multimodal cues, including language, audio, and vision, acknowledging instances where causes may not be exclusively represented in the textual data. Our proposed model for emotion-cause analysis is meticulously structured into three core segments: (i) embedding extraction, (ii) cause-pair extraction & emotion classification, and (iii) cause extraction using QA after finding pairs. Leveraging state-of-the-art techniques and fine-tuning on task-specific datasets, our model effectively unravels the intricate web of conversational dynamics and extracts subtle cues signifying causality in emotional expressions. Our team, AIMA, demonstrated strong performance in the SemEval-2024 Task 3 competition. We ranked as the 10th in subtask 1 and the 6th in subtask 2 out of 23 teams.

AIMA at SemEval-2024 Task 3: Simple Yet Powerful Emotion Cause Pair Analysis

TL;DR

This work tackles emotion-cause pair extraction in conversations through a text-centric three-component architecture: embedding extraction with EmoBERTa, cause-pair extraction plus emotion classification via a Transformer encoder that yields a causality matrix, and QA-driven extraction of cause spans after pair detection. Although designed to handle multimodal data, the study focuses on textual data for Subtask 1 and demonstrates competitive performance in SemEval-2024 Task 3, ranking 10th in Subtask 1 and 5th in Subtask 2 among 23 teams. Evaluations rely on F1-based metrics, including strict and proportional matching for textual spans, and emphasize the trade-off between emotion recognition accuracy and precise cause extraction. The results suggest that simple, efficient modeling can yield strong insights into conversational emotions while pointing to avenues for lightweight multimodal enhancements in future work.

Abstract

The SemEval-2024 Task 3 presents two subtasks focusing on emotion-cause pair extraction within conversational contexts. Subtask 1 revolves around the extraction of textual emotion-cause pairs, where causes are defined and annotated as textual spans within the conversation. Conversely, Subtask 2 extends the analysis to encompass multimodal cues, including language, audio, and vision, acknowledging instances where causes may not be exclusively represented in the textual data. Our proposed model for emotion-cause analysis is meticulously structured into three core segments: (i) embedding extraction, (ii) cause-pair extraction & emotion classification, and (iii) cause extraction using QA after finding pairs. Leveraging state-of-the-art techniques and fine-tuning on task-specific datasets, our model effectively unravels the intricate web of conversational dynamics and extracts subtle cues signifying causality in emotional expressions. Our team, AIMA, demonstrated strong performance in the SemEval-2024 Task 3 competition. We ranked as the 10th in subtask 1 and the 6th in subtask 2 out of 23 teams.
Paper Structure (11 sections, 3 figures, 3 tables)

This paper contains 11 sections, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. System Overview
  4. Experimental Setup
  5. Dataset Preparation
  6. Training
  7. Evaluation Metrics
  8. Results
  9. Quantitative Findings
  10. Error Analysis
  11. Conclusion

Figures (3)

  • Figure 1: An example of the annotated conversation in ECF wang2023multimodal dataset, illustrating the multimodal nature of emotion causes. Each arc points from the cause utterance to the emotion it triggers. The cause spans have been highlighted in yellow.
  • Figure 2: The schema of our proposed model for emotion-cause analysis, meticulously partitioned into three core segments: Embedding Extraction, Cause Pair Extraction and Emotion Classification, and Cause Extraction After Finding Pairs
  • Figure 3: An example of the model's question-answering mechanism in action. After classifying emotions in the dialogue and creating the causality matrix, a question prompt is generated only for detected emotion-cause pairs. This diagram demonstrates the process of identifying the causative segment within the dialogue that led to the emotional response, with the causative text being highlighted in the context of the detected pairs.