Aspect-Based Sentiment Analysis in Persian Using a Fine-Tuned LLaMA۳ Model

Aspect-Based Sentiment Analysis (ABSA) is less developed for Persian than for English, particularly for extracting complete aspect-category-opinion-sentiment structures rather than classifying sentiment for predefined aspects. This paper presents a generative Persian ABSA framework based on fine-tuned LLaMA۳-family instruction models. Given a Persian review, the model generates a JSON array of quadruples containing aspect terms, categories, opinion spans, and sentiment labels.The training data are drawn from a GPT-۳.۵-assisted Persian ABSA corpus whose annotations were reviewed and filtered by human annotators; we therefore treat it as GPT-assisted/human-validated rather than purely human-authored. To reduce circular evaluation, we also construct an independent human-only test set from raw Persian reviews. Three annotators manually annotated ۲۴۰ reviews from scratch, and a two-of-three consensus procedure produced ۵۵۵ gold aspect-level labels from ۲۳۹ reviews.We evaluate zero-shot LLaMA۳, fine-tuned LLaMA۳ models, and Dorna-LLaMA۳-۸B-Instruct as a Persian-oriented generative baseline under the same schema, parser, and metrics. On the human-only test set, zero-shot LLaMA۳ obtains ۰.۱۴۲۶ aspect F۱ and ۰.۱۱۸۵ aspect-sentiment F۱. Fine-tuned LLaMA۳-CG improves the five-seed mean scores to ۰.۵۴۷۷ ۰.۰۲۲۹ and ۰.۴۳۹۴, ۰.۰۲۱۲, while Dorna-CG reaches ۰.۵۸۰۹, ۰.۰۱۲۳ and ۰.۴۸۹۷, ۰.۰۱۹۴. Adding ParsiNLU substantially improves in-domain ParsiNLU performance but reduces human-only performance, revealing a domain and annotation-style trade-off.

کلیدواژه ها:

Aspect-Based Sentiment Analysis ، Persian Language ، Sentiment Analysis ، LLaMA۳ Model ، ChatGPT ، Natural Language Processing