Dialogue summarization (DS) plays a vital role in improving customer service efficiency by automatically generating concise summaries from lengthy multi-turn dialogues. However, existing studies largely overlook the fine-grained sentiment dynamics expressed by customers, and most DS datasets lack detailed sentiment annotations. These limitations hinder both accurate service quality assessment and the development of sentiment-aware summarization models. To address these challenges, we propose a three-stage approach to building an aspect-aware sentiment dataset, comprising: (1) aspect-anchored dialogue rewriting, (2) dialogue-anchored explainable label generation, and (3) label-dialogue integrated summarization. Building upon this scheme, we construct FOCUS, a Fine-grained customer-Oriented Chinese dialogUe Summarization dataset. FOCUS is the first Chinese dataset with 12,948 dialogues annotated for multi-level aspects, sentiment polarity, opinion content, emotions, as well as customer-oriented formatted and free-style sentiment summaries. To demonstrate the challenges and utility of FOCUS, we benchmark a range of summarization models on FOCUS and observe that current methods often exhibit misalignment between aspects and sentiments. Meanwhile, we find that a Chain-of-Thought approach can enhance faithfulness and interpretability, highlighting promising directions for future research on this dataset. FOCUS serves as a valuable resource to advance research in sentiment-aware DS and related tasks.

Textual Emotion-Cause Pair Extraction in Conversations (TECPE) requires identifying emotion utterances, their categories, and the textual cause spans that explain them. Although recent systems improve aggregate scores, clean supervision can induce evidence bias: models exploit explicit emotion cues, cause–emotion proximity, and speaker/discourse priors rather than evidence sufficient for causal explanation. We introduce SAAM-EC, a sufficiency-aware adversarial evidence masking framework for span-grounded TECPE. SAAM-EC adversarially masks high-impact conversational evidence units during training, but treats masking as a semantic intervention rather than a label-preserving perturbation. When masking removes the cause span or necessary bridge context, the original span may no longer be extractable or justified. SAAM-EC therefore combines adversarial masking with a sufficiency gate and an extractability guard, training masked examples to recover the original span only when the remaining dialogue still supports the emotion–cause relation and preserves the annotated span; otherwise, it trains the model with an unanswerability objective. All auxiliary components are discarded at inference, leaving the backbone unchanged. Experiments on span-level TECPE and utterance-level ECPE benchmarks show consistent performance gains over clean backbones and masking-based alternatives. Diagnostics further show better recovery on distant and cross-speaker causes and stronger abstention under destructive masks.