Heng-yang Lu , Tian-ci Liu , Rui Cong , Jun Yang , Qiang Gan , Wei Fang , Xiao-jun Wu
{"title":"QAIE:基于 LLM 的数量扩增和信息增强技术,适用于基于几个方面的情感分析","authors":"Heng-yang Lu , Tian-ci Liu , Rui Cong , Jun Yang , Qiang Gan , Wei Fang , Xiao-jun Wu","doi":"10.1016/j.ipm.2024.103917","DOIUrl":null,"url":null,"abstract":"<div><div>Aspect-based Sentiment Analysis (ABSA) aims to extract fine-grained sentiment information from online reviews. Few-shot ABSA faces challenges with limited labeled data and recent generative models have outperformed traditional classification models. Existing methods use Question Answering (QA) templates with Text-to-Text Transfer Transformer (T5) to extract sentiment elements, introducing a generative sentiment analysis paradigm. However, these models often fail to fully grasp ABSA rules, generating non-standard or incorrect outputs. This issue also arises with large language models (LLMs) due to insufficient labeled data for tuning and learning. Additionally, ABSA datasets often include many short, uninformative reviews, complicating sentiment element extraction in few-shot scenarios. This paper addresses two major challenges in few-shot ABSA: (1) <em>How to let the generative model well understand the ABSA rules under few-shot scenarios</em>. (2) <em>How to enhance the review text with richer information</em>. We propose a <strong>Q</strong>uantity <strong>A</strong>ugmentation and <strong>I</strong>nformation <strong>E</strong>nhancement (<strong>QAIE</strong>) approach, leveraging LLMs to generate fluent texts and infer implicit information. First, we propose a quantity augmentation module, which leverages the large language model (LLM) to obtain sufficient labeled data for the generative model to learn the ABSA rules better. Then, we introduce an information enhancement module, which brings more informative input to the generative model by enhancing the information in the review. Comprehensive experiments on five ABSA tasks using three widely-used datasets demonstrate that our QAIE model achieves approximately 10% improvement over state-of-the-art models. Specifically, for the most challenging ASQP task, our LLM-based model is compared with the existing state-of-the-art models on datasets Rest15 and Rest16, achieving F1 gains of 9.42% and 6.45% respectively in the <span><math><mrow><mi>k</mi><mo>=</mo><mn>5</mn></mrow></math></span> few-shot setting.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 1","pages":"Article 103917"},"PeriodicalIF":7.4000,"publicationDate":"2024-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"QAIE: LLM-based Quantity Augmentation and Information Enhancement for few-shot Aspect-Based Sentiment Analysis\",\"authors\":\"Heng-yang Lu , Tian-ci Liu , Rui Cong , Jun Yang , Qiang Gan , Wei Fang , Xiao-jun Wu\",\"doi\":\"10.1016/j.ipm.2024.103917\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><div>Aspect-based Sentiment Analysis (ABSA) aims to extract fine-grained sentiment information from online reviews. Few-shot ABSA faces challenges with limited labeled data and recent generative models have outperformed traditional classification models. Existing methods use Question Answering (QA) templates with Text-to-Text Transfer Transformer (T5) to extract sentiment elements, introducing a generative sentiment analysis paradigm. However, these models often fail to fully grasp ABSA rules, generating non-standard or incorrect outputs. This issue also arises with large language models (LLMs) due to insufficient labeled data for tuning and learning. Additionally, ABSA datasets often include many short, uninformative reviews, complicating sentiment element extraction in few-shot scenarios. This paper addresses two major challenges in few-shot ABSA: (1) <em>How to let the generative model well understand the ABSA rules under few-shot scenarios</em>. (2) <em>How to enhance the review text with richer information</em>. We propose a <strong>Q</strong>uantity <strong>A</strong>ugmentation and <strong>I</strong>nformation <strong>E</strong>nhancement (<strong>QAIE</strong>) approach, leveraging LLMs to generate fluent texts and infer implicit information. First, we propose a quantity augmentation module, which leverages the large language model (LLM) to obtain sufficient labeled data for the generative model to learn the ABSA rules better. Then, we introduce an information enhancement module, which brings more informative input to the generative model by enhancing the information in the review. Comprehensive experiments on five ABSA tasks using three widely-used datasets demonstrate that our QAIE model achieves approximately 10% improvement over state-of-the-art models. Specifically, for the most challenging ASQP task, our LLM-based model is compared with the existing state-of-the-art models on datasets Rest15 and Rest16, achieving F1 gains of 9.42% and 6.45% respectively in the <span><math><mrow><mi>k</mi><mo>=</mo><mn>5</mn></mrow></math></span> few-shot setting.</div></div>\",\"PeriodicalId\":50365,\"journal\":{\"name\":\"Information Processing & Management\",\"volume\":\"62 1\",\"pages\":\"Article 103917\"},\"PeriodicalIF\":7.4000,\"publicationDate\":\"2024-10-22\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Information Processing & Management\",\"FirstCategoryId\":\"94\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S0306457324002760\",\"RegionNum\":1,\"RegionCategory\":\"管理学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"COMPUTER SCIENCE, INFORMATION SYSTEMS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Information Processing & Management","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0306457324002760","RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}
QAIE: LLM-based Quantity Augmentation and Information Enhancement for few-shot Aspect-Based Sentiment Analysis
Aspect-based Sentiment Analysis (ABSA) aims to extract fine-grained sentiment information from online reviews. Few-shot ABSA faces challenges with limited labeled data and recent generative models have outperformed traditional classification models. Existing methods use Question Answering (QA) templates with Text-to-Text Transfer Transformer (T5) to extract sentiment elements, introducing a generative sentiment analysis paradigm. However, these models often fail to fully grasp ABSA rules, generating non-standard or incorrect outputs. This issue also arises with large language models (LLMs) due to insufficient labeled data for tuning and learning. Additionally, ABSA datasets often include many short, uninformative reviews, complicating sentiment element extraction in few-shot scenarios. This paper addresses two major challenges in few-shot ABSA: (1) How to let the generative model well understand the ABSA rules under few-shot scenarios. (2) How to enhance the review text with richer information. We propose a Quantity Augmentation and Information Enhancement (QAIE) approach, leveraging LLMs to generate fluent texts and infer implicit information. First, we propose a quantity augmentation module, which leverages the large language model (LLM) to obtain sufficient labeled data for the generative model to learn the ABSA rules better. Then, we introduce an information enhancement module, which brings more informative input to the generative model by enhancing the information in the review. Comprehensive experiments on five ABSA tasks using three widely-used datasets demonstrate that our QAIE model achieves approximately 10% improvement over state-of-the-art models. Specifically, for the most challenging ASQP task, our LLM-based model is compared with the existing state-of-the-art models on datasets Rest15 and Rest16, achieving F1 gains of 9.42% and 6.45% respectively in the few-shot setting.
期刊介绍:
Information Processing and Management is dedicated to publishing cutting-edge original research at the convergence of computing and information science. Our scope encompasses theory, methods, and applications across various domains, including advertising, business, health, information science, information technology marketing, and social computing.
We aim to cater to the interests of both primary researchers and practitioners by offering an effective platform for the timely dissemination of advanced and topical issues in this interdisciplinary field. The journal places particular emphasis on original research articles, research survey articles, research method articles, and articles addressing critical applications of research. Join us in advancing knowledge and innovation at the intersection of computing and information science.