Researchers have developed a new framework for extracting core opinions from multimodal and multilingual data streams, specifically for Science and Technology Intelligence (STI). This approach utilizes VideoLLaMA2 and VideoLLaMA2.1 as base models, fine-tuned using QLoRA on a dataset of 2,194 samples. The fine-tuned model demonstrates improved performance in generating structured JSON core-opinion outputs, significantly boosting F1-scores for Spanish and Russian language extraction compared to zero-shot settings. Additionally, a post-extraction triage module based on Fuzzy Cumulative Prospect Theory is incorporated to assess case-level value for downstream screening. AI
IMPACT This research could improve the efficiency and accuracy of information filtering in large-scale multimodal and multilingual datasets for intelligence analysis.
RANK_REASON The cluster contains an academic paper detailing a new approach to opinion extraction using fine-tuned LLMs. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Gotit.pub
- Hugging Face
- QLoRA
- Russian
- ScienceCast
- Spanish
- VideoLLaMA2
- VideoLLaMA2.1
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
