Learning to Extract Cross-Domain Aspects and Understanding Sentiments Using Large Language Models

Karukriti Kaushik Ghosh, Chiranjib Sur · 2025

Read Learning to Extract Cross-Domain Aspects and Understanding Sentiments Using Large Language Models by Karukriti Kaushik Ghosh, Chiranjib Sur (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-08974.txt.

Introduction

Karukriti Kaushik Ghosh and Chiranjib Sur apply large language models to fine-grained reading of customer reviews, in a 2025 paper. Simple sentiment scores say whether a review is glad or mad overall. Aspect-based analysis goes further: it pulls out each feature mentioned — quality, price, service — and judges the feeling aimed at each one.

That split view lets firms see exactly which parts of a product please or annoy buyers. The pipeline finds the aspects, sorts the feeling for each, and rolls the scores up for a whole review. The authors test how well language models handle the task across different product areas, where words shift meaning. A method that travels across domains saves firms from rebuilding their tools for each new product line.

Catalog record

CREATORKarukriti Kaushik Ghosh, Chiranjib Sur
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-08974

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Learning to Extract Cross-Domain Aspects and Understanding Sentiments Using Large Language Models Karukriti Kaushik Ghosh, Chiranjib Sur Abstract Aspect-based sentiment analysis (ASBA) is a refined approach to sentiment analysis that aims to extract and classify sentiments based on specific aspects or features of a product, service, or entity. Unlike traditional sentiment analysis, which assigns a general sentiment score to entire reviews or texts, ABSA focuses on breaking down the text into individual components or aspects (e.g., quality, price, service) and evaluating the sentiment towards each. This allows for a more granular level of understanding of customer opinions, enabling businesses to pinpoint specific areas of strength and improvement. The process involves several key steps, including aspect extraction, sentiment classification, and aspect-level sentiment aggregation for a review paragraph or any other form that the users have provided. ABSA has significant applications in areas such as product reviews, social media monitoring, customer feedback analysis, and market research. By leveraging techniques from natural language processing (NLP) and machine learning, ABSA facilitates the extraction of valuable insights, enabling companies to make data-driven decisions that enhance customer satisfaction and optimize offerings. As ABSA evolves, it holds the potential to greatly improve personalized customer experiences by providing a deeper understanding of sentiment across various product aspects. In this work, we have analyzed the strength of LLMs for a complete cross-domain aspect-based sentiment analysis with the aim of defining the framework for certain products and using it for other similar situations. We argue that it is possible to that at an effectiveness of 92% accuracy for the Aspect Based Sentiment Analysis dataset of SemEval-2015 Task 12. † † tocauthor: Ivar Ekeland, Roger Temam, Jeffrey Dean, David Grove, Craig Chambers, Kim B. Bruce, and Elisa Bertino † † institute: Indian Institute of Technology, Guwahati Assam 781039, India, 1 Introduction Figure 1: Diagram illustrating how specific knowledge source or knowledge graph combined with Large Language Models can help in understanding the specifics of a product and the sentiments of the reviews can be extracted with more precision. Aspect-based sentiment analysis (ABSA) tang2016aspect is an advanced technique within the broader field of sentiment analysis that goes beyond general sentiment classification. Unlike traditional sentiment analysis, which assigns an overall sentiment (positive, negative, or neutral) to a piece of text, ABSA breaks down the text into specific aspects or features related to the subject being discussed, such as product quality, customer service, pricing, or delivery jin2020bert . Each aspect is then evaluated for sentiment, offering a more nuanced and precise understanding of customer opinions jin2020bert . The growing reliance on online reviews, social media discussions, and customer feedback has made ABSA an invaluable tool for businesses seeking to understand consumer sentiment at a deeper level blitzer2007biographies . By analyzing sentiments tied to individual aspects of a product or service, companies can gain insights into what aspects of their offerings are resonating with customers, which elements require improvement, and how different features contribute to overall satisfaction blitzer2007biographies . This allows businesses to make targeted improvements to specific aspects of their products or services rather than making broad changes based on generalized feedback dodge2020fine . ABSA has applications across various domains, including e-commerce, hospitality, healthcare, and social media monitoring, where understanding the sentiment surrounding individual aspects of a product or service can lead to more informed decision-making elsayed2018large . Leveraging natural language processing (NLP) and machine learning techniques, ABSA enables automated and scalable khosla2020supervised sentiment analysis, helping businesses process large volumes of data quickly and efficiently elsayed2018large . As ABSA continues to evolve, it holds the potential to significantly enhance personalized customer experiences and improve customer satisfaction by providing businesses with actionable, aspect-level insights ganin2016domain , gunel2020supervised , johnson2018training . The overall structure of this document is organized as Section 1 with introduction and literature review, Section 2 has the problem statement, Section 3 has the architectural details, Section 4 has a brief methodology, Section 5 showcases the experimental setup and results, and Section 6 details the conclusion and future work. The main contribution of this work: 1) We proposed a framework for cross-domain aspect-based sentiment analysis 2) We modified the existing Aspect Based Sentiment Analysis dataset of SemEval-2015 Task 12 so that we can adopt it for cross-domain sentiment analysis through the use of external knowledge sources and data leverages. 3) We proposed a model that will be more business-friendly as this will require less training and expertise in data science skills 4) We demonstrated that cross-domain does not require finetuning, which is costly mainly to fine-tune a big model like BERT or other Large Language Models. 2 Problem Statement In the age of digital transformation, businesses are increasingly relying on customer feedback kenton2019bert from various online platforms such as product reviews, social media, and customer support forums to gauge public perception and improve their products or services. While traditional sentiment analysis methods can provide a broad understanding of whether customers are generally satisfied or dissatisfied, they fail to capture the nuanced opinions expressed about specific aspects lewis2019bart or features of a product or service. This presents a significant challenge, as businesses need detailed insights into which particular elements of their offerings are positively or negatively perceived by customers. The problem lies in the need for a more granular approach to sentiment analysis—Aspect-Based Sentiment Analysis (ABSA). ABSA allows organizations to extract, identify, and classify sentiments regarding specific aspects or features of a product, service, or entity. For example, in an e-commerce setting, customers may express satisfaction with the quality of a product but dissatisfaction with its price or delivery time. Traditional sentiment analysis would label the review as positive or negative overall, failing to highlight these critical distinctions lewis2019bart . The key challenge in ABSA is to accurately identify relevant aspects within textual data and classify sentiments towards these aspects in a manner that reflects the context of the review or feedback. This requires advanced techniques in natural language processing (NLP) khandelwal2020nearest and machine learning, which can address issues such as ambiguity, aspect extraction, and sentiment classification. Additionally, the ability to scale ABSA models to handle large volumes of diverse data while maintaining high accuracy is a significant hurdle khandelwal2020nearest . By overcoming these challenges, cross-domain ABSA models can provide businesses with a deeper, more actionable understanding of customer opinions and focus on improving them. It can help businesses identify strengths and weaknesses in specific areas, such as product features, customer service, or pricing, leading to more targeted improvements and a better alignment of offerings with customer expectations. Furthermore, ABSA can enhance personalized marketing strategies, improve customer satisfaction, and foster greater customer loyalty by addressing the issues that matter most to individual customers. Therefore, the problem at hand is to develop an effective and scalable Aspect-Based Sentiment Analysis system that can accurately extract and assess sentiment at the aspect level, providing businesses with the insights they need to optimize their products, services, and customer experiences. 3 Architecture The first part of the architecture deals with understanding what an aspect is and how it varies from product to product and domain to domain. To counter that problem, we have used LLMs (like llama) with external information for better understanding. We can use any model like

Read the complete text