Overview
The ganchengguang/USA-7B-instruction-incontext-learning model is a 7 billion parameter language model primarily focused on Japanese sentiment analysis. Developed by Chengguang Gan, Qinghao Zhang, and Tatsunori Mori, this model is based on the research presented in their paper "USA: Universal Sentiment Analysis Model & Construction of Japanese Sentiment Text Classification and Part of Speech Dataset" (arXiv:2309.03787).
Key Capabilities
- Japanese Sentiment Analysis: Designed to classify the overall sentiment of Japanese text as positive, negative, or neutral.
- Noun-level Sentiment Identification: Beyond overall sentiment, it can identify and list specific nouns within the text and assign them a sentiment label (positive, neutral, negative).
- In-Context Learning (ICL): Leverages in-context examples to guide its predictions, enhancing its ability to understand and apply sentiment rules.
- Instruction Learning: Incorporates explicit instructions to perform sentiment analysis tasks.
- Unifine Format: Utilizes a specific "Unifine" format for both input and output, which combines in-context learning, instruction learning, and formatted text for consistent processing.
Good For
- Japanese Text Sentiment Classification: Ideal for applications requiring sentiment analysis of Japanese language content.
- Detailed Sentiment Extraction: Useful for identifying specific positive, neutral, or negative entities (nouns) within Japanese sentences.
- Research in Japanese NLP: Provides a specialized tool for researchers working on sentiment analysis and part-of-speech tagging in Japanese.