Orkhan/llama-2-7b-absa: Aspect-Based Sentiment Analysis Model
This model is a fine-tuned version of the 7 billion parameter Llama-2 base model, specifically optimized by Orkhan for Aspect-Based Sentiment Analysis (ABSA). It leverages a manually labeled dataset of 2000 sentences to accurately extract aspects, associated opinions, and their sentiments from text.
Key Capabilities
- Detailed Sentiment Analysis: Identifies specific aspects within a sentence (e.g., "weather", "smell"), the opinions expressed about them (e.g., "nice", "bad"), and their corresponding sentiments (e.g., "Positive", "Negative").
- Generalized ABSA: Offers an advantage over traditional ABSA models by generalizing well across domains, potentially reducing the need for extensive domain-specific labeled data for training.
- Output Structure: Provides structured output including lists of aspects, opinions, sentiments, and combined phrases (e.g., "nice weather").
Usage Notes
- The model is trained on sentences, not paragraphs, which should be considered during inference.
- It is designed to run efficiently on T4-GPU-enabled environments, such as free Google Colab notebooks.
- Includes helper functions for processing prompts and structuring the model's output for easier interpretation.
This model is ideal for applications requiring granular sentiment insights beyond overall text sentiment, enabling a deeper understanding of user feedback or reviews.