Dynamic Span Interaction and Graph-Aware Memory for Entity-Level Sentiment Classification

2025-09-15

AI Summary

Entity-level sentiment classification requires modeling fine-grained interactions between target entities and distant, cross-sentence sentiment expressions, while addressing linguistic complexities such as coreference ambiguity, negation, and overlapping opinions. To this end, we propose SpanEIT—a novel framework integrating a dynamic span interaction mechanism, a graph attention network (GAT), and a coreference-aware memory module. This design enables context-adaptive representation learning for entity spans, syntax- and co-occurrence-guided relational reasoning, and cross-sentence consistency constraints on entity sentiment predictions. SpanEIT achieves significant improvements over state-of-the-art methods on FSAD, BARU, and IMDB benchmarks, with notable gains in both accuracy and F1 score. Ablation studies and interpretability analyses confirm the individual efficacy and synergistic contributions of each component.

Technology Category

• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Sentiment analysis and opinion mining

Abstract

Entity-level sentiment classification involves identifying the sentiment polarity linked to specific entities within text. This task poses several challenges: effectively modeling the subtle and complex interactions between entities and their surrounding sentiment expressions; capturing dependencies that may span across sentences; and ensuring consistent sentiment predictions for multiple mentions of the same entity through coreference resolution. Additionally, linguistic phenomena such as negation, ambiguity, and overlapping opinions further complicate the analysis. These complexities make entity-level sentiment classification a difficult problem, especially in real-world, noisy textual data. To address these issues, we propose SpanEIT, a novel framework integrating dynamic span interaction and graph-aware memory mechanisms for enhanced entity-sentiment relational modeling. SpanEIT builds span-based representations for entities and candidate sentiment phrases, employs bidirectional attention for fine-grained interactions, and uses a graph attention network to capture syntactic and co-occurrence relations. A coreference-aware memory module ensures entity-level consistency across documents. Experiments on FSAD, BARU, and IMDB datasets show SpanEIT outperforms state-of-the-art transformer and hybrid baselines in accuracy and F1 scores. Ablation and interpretability analyses validate the effectiveness of our approach, underscoring its potential for fine-grained sentiment analysis in applications like social media monitoring and customer feedback analysis.

Problem

• Modeling entity-sentiment interactions across sentences
• Ensuring consistent sentiment for coreferenced entities
• Handling linguistic complexities like negation and ambiguity

Innovation

• Dynamic span interaction for entity-sentiment modeling
• Graph attention network capturing syntactic relations
• Coreference-aware memory ensuring entity-level consistency