{"title":"基于影响力评分的图像情感调色板推荐,用于图像广告","authors":"Juhee Han, Younghoon Lee","doi":"10.1007/s10660-024-09851-4","DOIUrl":null,"url":null,"abstract":"<p>As image-based communication proliferates, the business value of image sentiment analysis is rapidly growing, particularly in fields like advertising where consumers receive emotional cues through visual stimuli. However, most existing research on image sentiment analysis has focused more on developing sentiment classification models rather than exploring specific factors contributing to image sentiment. Therefore, this study proposes a methodology for extracting color palettes to represent image sentiments, emphasizing the significance of color palettes as highlighted in various studies. Previous approaches to color palette extraction have included heuristic methods, survey-based selection, or utilizing clustering algorithms like K-means clustering based on color frequencies in images. In this study, we calculate the influence scores of colors for classifying image sentiments and propose deriving representative sentiment-color palettes based on these scores. Initially, we train a multi-label classification model to predict the sentiment labels of images and then create datasets for distorted images where pixels corresponding to specific colors are removed. By comparing the model outputs obtained from these distorted images with the original dataset, we obtain quantitative influence scores of colors for classifying sentiment labels. Furthermore, we extract sentiment-color palettes consisting of four important colors for 30 different sentiments. Experimental results demonstrate higher evaluation scores compared to previous studies.</p>","PeriodicalId":47264,"journal":{"name":"Electronic Commerce Research","volume":null,"pages":null},"PeriodicalIF":3.7000,"publicationDate":"2024-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Image sentiment considering color palette recommendations based on influence scores for image advertisement\",\"authors\":\"Juhee Han, Younghoon Lee\",\"doi\":\"10.1007/s10660-024-09851-4\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p>As image-based communication proliferates, the business value of image sentiment analysis is rapidly growing, particularly in fields like advertising where consumers receive emotional cues through visual stimuli. However, most existing research on image sentiment analysis has focused more on developing sentiment classification models rather than exploring specific factors contributing to image sentiment. Therefore, this study proposes a methodology for extracting color palettes to represent image sentiments, emphasizing the significance of color palettes as highlighted in various studies. Previous approaches to color palette extraction have included heuristic methods, survey-based selection, or utilizing clustering algorithms like K-means clustering based on color frequencies in images. In this study, we calculate the influence scores of colors for classifying image sentiments and propose deriving representative sentiment-color palettes based on these scores. Initially, we train a multi-label classification model to predict the sentiment labels of images and then create datasets for distorted images where pixels corresponding to specific colors are removed. By comparing the model outputs obtained from these distorted images with the original dataset, we obtain quantitative influence scores of colors for classifying sentiment labels. Furthermore, we extract sentiment-color palettes consisting of four important colors for 30 different sentiments. Experimental results demonstrate higher evaluation scores compared to previous studies.</p>\",\"PeriodicalId\":47264,\"journal\":{\"name\":\"Electronic Commerce Research\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":3.7000,\"publicationDate\":\"2024-05-09\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Electronic Commerce Research\",\"FirstCategoryId\":\"91\",\"ListUrlMain\":\"https://doi.org/10.1007/s10660-024-09851-4\",\"RegionNum\":4,\"RegionCategory\":\"管理学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"BUSINESS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Electronic Commerce Research","FirstCategoryId":"91","ListUrlMain":"https://doi.org/10.1007/s10660-024-09851-4","RegionNum":4,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"BUSINESS","Score":null,"Total":0}
Image sentiment considering color palette recommendations based on influence scores for image advertisement
As image-based communication proliferates, the business value of image sentiment analysis is rapidly growing, particularly in fields like advertising where consumers receive emotional cues through visual stimuli. However, most existing research on image sentiment analysis has focused more on developing sentiment classification models rather than exploring specific factors contributing to image sentiment. Therefore, this study proposes a methodology for extracting color palettes to represent image sentiments, emphasizing the significance of color palettes as highlighted in various studies. Previous approaches to color palette extraction have included heuristic methods, survey-based selection, or utilizing clustering algorithms like K-means clustering based on color frequencies in images. In this study, we calculate the influence scores of colors for classifying image sentiments and propose deriving representative sentiment-color palettes based on these scores. Initially, we train a multi-label classification model to predict the sentiment labels of images and then create datasets for distorted images where pixels corresponding to specific colors are removed. By comparing the model outputs obtained from these distorted images with the original dataset, we obtain quantitative influence scores of colors for classifying sentiment labels. Furthermore, we extract sentiment-color palettes consisting of four important colors for 30 different sentiments. Experimental results demonstrate higher evaluation scores compared to previous studies.
期刊介绍:
The Internet and the World Wide Web have brought a fundamental change in the way that individuals access data, information and services. Individuals have access to vast amounts of data, to experts and services that are not limited in time or space. This has forced business to change the way in which they conduct their commercial transactions with their end customers and with other businesses, resulting in the development of a global market through the Internet. The emergence of the Internet and electronic commerce raises many new research issues. The Electronic Commerce Research journal will serve as a forum for stimulating and disseminating research into all facets of electronic commerce - from research into core enabling technologies to work on assessing and understanding the implications of these technologies on societies, economies, businesses and individuals. The journal concentrates on theoretical as well as empirical research that leads to better understanding of electronic commerce and its implications. Topics covered by the journal include, but are not restricted to the following subjects as they relate to the Internet and electronic commerce: Dissemination of services through the Internet;Intelligent agents technologies and their impact;The global impact of electronic commerce;The economics of electronic commerce;Fraud reduction on the Internet;Mobile electronic commerce;Virtual electronic commerce systems;Application of computer and communication technologies to electronic commerce;Electronic market mechanisms and their impact;Auctioning over the Internet;Business models of Internet based companies;Service creation and provisioning;The job market created by the Internet and electronic commerce;Security, privacy, authorization and authentication of users and transactions on the Internet;Electronic data interc hange over the Internet;Electronic payment systems and electronic funds transfer;The impact of electronic commerce on organizational structures and processes;Supply chain management through the Internet;Marketing on the Internet;User adaptive advertisement;Standards in electronic commerce and their analysis;Metrics, measurement and prediction of user activity;On-line stock markets and financial trading;User devices for accessing the Internet and conducting electronic transactions;Efficient search techniques and engines on the WWW;Web based languages (e.g., HTML, XML, VRML, Java);Multimedia storage and distribution;Internet;Collaborative learning, gaming and work;Presentation page design techniques and tools;Virtual reality on the net and 3D visualization;Browsers and user interfaces;Web site management techniques and tools;Managing middleware to support electronic commerce;Web based education, and training;Electronic journals and publishing on the Internet;Legal issues, taxation and property rights;Modeling and design of networks to support Internet applications;Modeling, design and sizing of web site servers;Reliability of intensive on-line applications;Pervasive devices and pervasive computing in electronic commerce;Workflow for electronic commerce applications;Coordination technologies for electronic commerce;Personalization and mass customization technologies;Marketing and customer relationship management in electronic commerce;Service creation and provisioning. Audience: Academics and professionals involved in electronic commerce research and the application and use of the Internet. Managers, consultants, decision-makers and developers who value the use of electronic com merce research results. Special Issues: Electronic Commerce Research publishes from time to time a special issue of the devoted to a single subject area. If interested in serving as a guest editor for a special issue, please contact the Editor-in-Chief J. Christopher Westland at westland@uic.edu with a proposal for the special issue. Officially cited as: Electron Commer Res