Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Extensible Multi-Granularity Fusion Network and Transferable Curriculum Learning for Aspect-based Sentiment Analysis

Xinran Li, Xiaowei Zhao, Yubo Zhu, Zhiheng Zhang, Zhiqi Huang, Hongkun Song, Jinglu Hu, Xinze Che, Yifan Lyu, Yong Zhou, Xiujuan Xu

TL;DR

This paper tackles fine-grained ABSA by introducing EMGF, a scalable framework that fuses dependency syntax, constituent syntax, semantic attention, and external knowledge through extensible EMSF blocks enhanced by multi-anchor triplet learning and orthogonal projection. It further introduces a task-specific curriculum learning strategy with five difficulty indicators to guide training from easy to hard, improving generalization. Experiments on multiple ABSA datasets show EMGF+CL achieving state-of-the-art performance on several benchmarks, with thorough ablations validating the contributions of each component. The approach offers a practical, extensible pathway for integrating diverse linguistic features without prohibitive computation, benefiting real-world ABSA systems.

Abstract

Aspect-based Sentiment Analysis (ABSA) aims to determine sentiment polarity toward specific aspects in text. Existing methods enrich semantic and syntactic representations through external knowledge or GNNs, but the growing diversity of linguistic features increases model complexity and lacks a unified, extensible framework. We propose an Extensible Multi-Granularity Fusion Network (EMGF) that integrates dependency syntax, constituent syntax, attention-based semantics, and external knowledge graphs. EMGF employs multi-anchor triplet learning and orthogonal projection to effectively fuse multi-granularity features and strengthen their interactions without additional computational overhead. Furthermore, we introduce the first task-specific curriculum learning framework for text-only ABSA, which assigns difficulty scores using five indicators and trains the model from easy to hard to mimic human learning and improve generalization. Experiments on SemEval 2014, Twitter, and MAMS datasets show that EMGF+CL consistently outperforms state-of-the-art ABSA models.

Extensible Multi-Granularity Fusion Network and Transferable Curriculum Learning for Aspect-based Sentiment Analysis

TL;DR

This paper tackles fine-grained ABSA by introducing EMGF, a scalable framework that fuses dependency syntax, constituent syntax, semantic attention, and external knowledge through extensible EMSF blocks enhanced by multi-anchor triplet learning and orthogonal projection. It further introduces a task-specific curriculum learning strategy with five difficulty indicators to guide training from easy to hard, improving generalization. Experiments on multiple ABSA datasets show EMGF+CL achieving state-of-the-art performance on several benchmarks, with thorough ablations validating the contributions of each component. The approach offers a practical, extensible pathway for integrating diverse linguistic features without prohibitive computation, benefiting real-world ABSA systems.

Abstract

Aspect-based Sentiment Analysis (ABSA) aims to determine sentiment polarity toward specific aspects in text. Existing methods enrich semantic and syntactic representations through external knowledge or GNNs, but the growing diversity of linguistic features increases model complexity and lacks a unified, extensible framework. We propose an Extensible Multi-Granularity Fusion Network (EMGF) that integrates dependency syntax, constituent syntax, attention-based semantics, and external knowledge graphs. EMGF employs multi-anchor triplet learning and orthogonal projection to effectively fuse multi-granularity features and strengthen their interactions without additional computational overhead. Furthermore, we introduce the first task-specific curriculum learning framework for text-only ABSA, which assigns difficulty scores using five indicators and trains the model from easy to hard to mimic human learning and improve generalization. Experiments on SemEval 2014, Twitter, and MAMS datasets show that EMGF+CL consistently outperforms state-of-the-art ABSA models.
Paper Structure (29 sections, 17 equations, 4 figures, 4 tables)

This paper contains 29 sections, 17 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Text Encoding Module
  5. Dependency GCN
  6. Constituent GCN
  7. Semantic GCN
  8. External Knowledge
  9. Extensible Multi-Stage Fusion Module
  10. Preprocessing Stage
  11. Multi-Anchor Triplet Learning
  12. Orthogonal Projection Techniques
  13. Fusion Stage
  14. Model Training
  15. Softmax Classifier
  16. ...and 14 more sections

Figures (4)

  • Figure 1: The sentence from a laptop review shows its dependency and constituent trees, containing two aspects with opposite sentiment polarities.
  • Figure 2: The overall architecture of our EMGF model.
  • Figure 3: The impact of the number of GCN and EMSF block on Restaurant dataset.
  • Figure 4: The impact of different $k$ on Restaurant dataset.