To address the valence and arousal score prediction task in Dimensional Aspect-Based Sentiment Analysis (DimABSA), we propose a two-stage strategy. In the first stage, we conduct post-training on a Large Language Model (LLM) via a Supervised Fine-Tuning (SFT) scheme, followed by generating initial predictions for valence and arousal scores. In the second stage, we perform distribution adaptation on the initial results by leveraging the training set distribution through various techniques, including Gaussian distribution modeling, quantile mapping, and the Sinkhorn algorithm.

This paper describes our system used in the SemEval-2025 Task 11: Bridging the Gap in Text-Based Emotion Detection. To address the highly subjective nature of emotion detection tasks, we propose a model ensemble strategy designed to capture the varying subjective perceptions of different users towards textual content. The base models of this ensemble strategy consist of several large language models, which are then combined using methods such as neural networks, decision trees, linear regression, and weighted voting. In Track A, out of 28 languages, our system achieved first place in 19 languages. In Track B, out of 11 languages, our system ranked first in 10 languages. Furthermore, our system attained the highest average performance across all languages in both Track A and Track B.