Reproduce “TCDA: Thread-Constrained Discourse-Aware Modeling for Conversational Sentiment Quadruple Analysis” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xinran Li, Xinze Che, Yifan Lyu, Zhiqi Huang, Xiujuan Xu
The TCDA framework refines Conversational Sentiment Quadruple Analysis (DiaASQ) by introducing a Thread-Constrained Directed Acyclic Graph (TC-DAG) for structural modeling and a Discourse-Aware Rotary Position Embedding (D-RoPE) to handle multi-scale positional dependencies. It achieved Micro F1 scores of 44.35 on the Chinese dataset and 39.69 on the English dataset for quadruple identification, showing competitive performance.
Key insights
- Explicitly modeling conversational threads using a Thread-Constrained Directed Acyclic Graph (TC-DAG) effectively suppresses structural noise from irrelevant branches while maintaining logical coherence within dialogues.
- Decomposing positional encoding into parallel token-level (Micro-RoPE) and utterance-level (Macro-RoPE) streams, as done in D-RoPE, successfully mitigates "Distance Dilution" by adapting to both local syntax and global discourse progression.
- The integration of structural (TC-DAG) and positional (D-RoPE) awareness provides complementary benefits, leading to a more comprehensive understanding of conversational context for fine-grained sentiment analysis.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.