AspectRAG: An Architecture of Retrieval and Generation for Aspect-Based Sentiment Analysis

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Natural Language Processing · Depth: Expert, quick

Summary

AspectRAG is a novel Retrieval and Generation architecture designed for Aspect-Based Sentiment Analysis (ASTE) in Portuguese, operating without supervised training. This method utilizes a Large Language Model (LLM) to extract aspects, which are then encoded into dense vectors. These vectors facilitate highly specific evidence retrieval through approximate search and ranking fusion. The retrieved evidence subsequently forms the context for a generator model, which produces the final sentiment triples. AspectRAG achieved significant performance on the ReLi and ReHol datasets, with scores up to 93.47% in ATE, 80.68% in OTE, and 69.83% in ASTE. These results surpass existing supervised models for Portuguese, including OTE-MTL, CMLA-MTL, and BOTE. An ablation study confirmed that aspect-guided semantic retrieval is the primary driver of these performance gains, with LLM size having a lesser impact.

Key takeaway

For AI Engineers developing sentiment analysis solutions for Portuguese, AspectRAG demonstrates that an unsupervised Retrieval and Generation architecture can achieve superior performance compared to supervised models. You should investigate integrating aspect-guided semantic retrieval into your systems, as it proved to be the main factor for performance gains, potentially reducing the need for extensive fine-tuning and labeled datasets.

Key insights

AspectRAG leverages aspect-guided retrieval and generation for unsupervised, high-performance sentiment analysis in Portuguese.

Principles

• Aspect-guided retrieval significantly boosts ASTE performance.
• Unsupervised RAG can outperform supervised models in specific tasks.

Method

AspectRAG extracts aspects via an LLM, encodes them as dense vectors, retrieves evidence using approximate search and ranking fusion, then generates sentiment triples from the retrieved context.

In practice

• Implement aspect-guided semantic retrieval for ASTE.
• Consider RAG architectures for unsupervised sentiment analysis.

Topics

• Aspect-Based Sentiment Analysis
• Retrieval-Augmented Generation
• Large Language Models
• Portuguese NLP
• Sentiment Triplet Extraction

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Aspect-Based Sentiment Analysis in the Accommodation Domain Using the BERTimbau Model — BERTimbau model fine-tuning strategies for Portuguese ABSA show trade-offs between recall and precision.
• Structured Summaries for Retrieval-Augmented Generation in Portuguese-Language Consumer Complaints — Structured summaries significantly improve dense retrieval effectiveness in RAG systems for noisy, lengthy texts.
• Zero-Shot to Full-Resource: Cross-lingual Transfer Strategies for Aspect-Based Sentiment Analysis — Cross-lingual transfer strategies for ABSA vary in effectiveness based on model architecture and resource availability.
• Vespa AI and Surpassing the Limits of Vector Search — Vector similarity alone is insufficient for robust search relevance; tensor-based retrieval offers superior flexibility and efficiency.
• DiffuSent: Towards a Unified Diffusion Framework for Aspect-Based Sentiment Analysis — DiffuSent unifies ABSA subtasks via non-auto-regressive boundary denoising, improving accuracy and speed for multi-word sentiment extraction.
• RAG Questions Need Parsing Too: Turn the User’s String Into Briefs for Retrieval and Generation — Parsing user questions into structured briefs for distinct retrieval and generation stages is crucial for RAG accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.