This system paper describes our participation in the SemEval-2025 Task-7 “Everyday Knowledge Across Diverse Languages and Cultures”. We attended two subtasks, i.e., Track 1: Short Answer Questions (SAQ), and Track 2: Multiple-Choice Questions (MCQ).The methods we used are retrieval augmented generation (RAGs) with open-sourced smaller LLMs (OS-sLLMs). To better adapt to this shared task, we created our own culturally aware knowledge base (CulKBs) by extracting Wikipedia content using keyword lists we prepared. We extracted both culturally-aware wiki-text and country-specific wiki-summary. In addition to the local CulKBs, we also have one system integrating live online search output via DuckDuckGo.Towards better privacy and sustainability, we aimed to deploy smaller LLMs (sLLMs) that are open-sourced on the Ollama platform.We share the prompts we developed using refinement techniques and report the learning curve of such prompts.The tested languages are English, Spanish, and Chinese for both tracks.Our resources and codes are shared via \url{https://github.com/aaronlifenghan/FLANS-2026}

We present our system for SemEval-2026 Task 3 on dimensional aspect-based sentiment regression. Our approach combines a hybrid RoBERTa encoder, which jointly predicts sentiment using regression and discretized classification heads, with large language models (LLMs) via prediction-level ensemble learning. The hybrid encoder improves prediction stability by combining continuous and discretized sentiment representations. We further explore in-context learning with LLMs and ridge-regression stacking to combine encoder and LLM predictions. Experimental results on the development set show that ensemble learning significantly improves performance over individual models, achieving substantial reductions in RMSE and improvements in correlation scores. Our findings demonstrate the complementary strengths of encoder-based and LLM-based approaches for dimensional sentiment analysis.Our development code and resources will be shared at \url{https://github.com/aaronlifenghan/ABSentiment}

For individuals in financial distress, understanding debt collection letters is critical. These documents are often unstructured, use complex legal language, and contain highly sensitive personal data. Automating information extraction is essential for assisting caseworkers, who currently perform this task manually; a slow and error-prone process. The sensitive nature of this data requires efficient, privacy-preserving, locally-deployed solutions. This paper compares the feasibility of various local NLP models for this task. We evaluated a feature-engineered Conditional Random Field (CRF), a fine-tuned spaCy NER model, and several Large Language Models (LLMs) (1.1B to 14B parameters) on a new synthetic dataset of 1,000 Dutch debt letters. Models were compared using accuracy (F1-score) and deployment metrics (CPU runtime, memory usage). Our results show a clear performance-resource trade-off. Lightweight CRF and spaCy models efficiently extracted structured data but failed in many critical unstructured fields. In contrast, LLM performance scaled directly with model size. The 14B DeepSeek model achieved the highest accuracy (95.2% average F1), successfully handling all field types. In conclusion, larger local LLMs are the most viable solution for accurate, private document processing. Alternatively, a hybrid approach using lightweight models for structured data and LLMs only for complex, unstructured fields, would also be adequate.