Due to their extensive common-sense knowledge and linguistic understanding, large language models (LLMs) have demonstrated remarkable capabilities in text comprehension and logical reasoning for natural language processing tasks. Traditional prompt-based learning methods, which rely on contextual pattern matching, have proven to be effective in extracting knowledge from LLMs. However, these approaches are constrained by training data pattern matching, overlook reasoning processes, and consequently suffer from suboptimal prompt performance and limited interpretability. Moreover, considering that the intermediate steps generated by single-chain reasoning may not effectively assist LLMs in identifying the sentiment polarity of aspect terms, and that multiple reasoning paths often exist for complex reasoning tasks to reach correct conclusions, this paper proposes a Multi-Chain Thought Prompt Learning framework (MT-CPL). Starting from fundamental concepts, this method simulates human multi-path reasoning patterns to progressively construct comprehensive thought processes and deeply explore sentiment cues. Based on syntactic structures and the semantic logic of text, the framework incorporates four distinct perspectives of text comprehension: hierarchical reading, experiential reading, keyword-based reading, and analogical reading. It establishes a multi-chain prompt template and employs voting mechanisms to select correct reasoning path outcomes. The MT-CPL approach aims to guide LLMs in mining multi-dimensional textual information from different perspectives, gradually uncovering hidden contextual sentiment clues, while mitigating issues caused by irrelevant sentiment cues in intermediate reasoning steps. By decomposing main tasks incrementally, the method achieves progressive reasoning, effectively reduces the difficulty of direct analysis, and further enhances model interpretability through the integration of inherent common-sense knowledge.

Multimodal Emotion Recognition in Conversation (MERC) aims to infer speakers’ affective states from heterogeneous cues within evolving dialogue contexts. Although graph-based methods have shown strong potential in modeling conversational structures, they often introduce history through dense or predefined connections and obtain unified multimodal representations before prediction. This paradigm may propagate irrelevant contextual information, overlook the different roles of same-speaker and cross-speaker dependencies, and underexploit useful corrective cues carried by inconsistent modalities. To address these limitations, we propose HiRoC, a multimodal emotion recognition framework that selectively routes emotion-relevant history and calibrates predictions using cross-modal disagreement. Specifically, a Historical Relevance Filter (HRF) scores how relevant each past utterance is to the current one and suppresses weakly related context before propagation. Guided by the resulting dependency prior, a Speaker-Typed Routing Graph (STRG) keeps intra- and inter-speaker dependencies apart and propagates the retained history along sparse, temporally directed paths. In parallel, the filtered modality-specific representations feed a hypergraph backbone for high-order utterance- and modality-level interactions, which the routing branch complements with speaker-aware temporal dependencies. A Cross-Modal Residual Correction (CMRC) module then uses modal disagreement to adjust the main prediction at the decision level, so that conflicting cues are not averaged away. On IEMOCAP and MELD, HiRoC outperforms representative state-of-the-art methods.

Document-level relation extraction (DocRE) finds relations across a whole document. It often needs evidence from several sentences. It also needs to link repeated entity mentions and follow multi-hop clues. Many large language model (LLM) based DocRE methods use fixed relation descriptions. They make little use of confident errors or samples with missing labels. We propose SPO-PA, which combines self-correcting prompt optimization (SPO) and preference alignment (PA). SPO compares LLM predictions with reference annotations. It then revises relation descriptions, clarifies relation boundaries, and adds role constraints. PA uses false negative (FN) conflicts as the main supervision signal. It also tests false positive (FP) conflicts under similar constraints. PA applies source-specific constraints to these signals. The constraints help smaller DocRE models separate correct relations from wrong candidates. On Re-DocRED, SPO improves F1 to 26.23% and Precision to 41.93%. It also achieves 37.60% F1 on HacRED. Both are the best LLM-only scores in our comparison and reproduce on a second, more recent backbone, though both stay far below supervised encoders; the extraction stage supplies conflict signals rather than final predictions. Among the PA variants, \(\text {PA}_{FN}\) gives the most stable gains on ATLOP, KD-DocRE, DREEAM, and DAATF. The largest gain is 2.89%.