Researchers have developed a new framework for extracting core opinions from multimodal and multilingual data, specifically for Science and Technology Intelligence (STI). This approach utilizes VideoLLaMA2 and VideoLLaMA2.1 models fine-tuned with QLoRA on a dataset of 2,194 samples. The fine-tuned VL2.1 model demonstrated improved performance in generating structured JSON outputs, significantly boosting F1-scores for Spanish and Russian language opinion extraction compared to its zero-shot capabilities. The framework also includes a triage module based on Fuzzy Cumulative Prospect Theory to assess case-level value for downstream screening. AI
IMPACT This research could improve the efficiency and accuracy of information filtering and analysis in large-scale STI contexts.
RANK_REASON The cluster describes a research paper detailing a new fine-tuning approach for multimodal and multilingual opinion extraction using LLMs.
Read on Hugging Face Daily Papers →
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Gotit.pub
- Hugging Face
- QLoRA
- Russian
- ScienceCast
- Spanish
- VideoLLaMA2
- VideoLLaMA2.1
- Fuzzy Cumulative Prospect Theory
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
