| 基于社会标签的图像情感分类标注研究 |
| |
| 引用本文： | 宋灵超,黄崑.基于社会标签的图像情感分类标注研究[J].图书情报工作,2016,60(21):103-112. |
| |
| 作者姓名： | 宋灵超 黄崑 |
| |
| 作者单位： | 1. 南开大学图书馆 天津 300350;
2. 北京师范大学政府管理学院 北京 100875 |
| |
| 基金项目： | 本文系教育部人文社会科学研究一般项目（青年基金项目）“基于大众参与的图像情感特征标引机制与方法研究”（项目编号：11YJC870010）研究成果之一。 |
| |
| 摘 要： | 目的/意义] 提出利用社会标签自动分类图片情感类型的方法，服务基于情感特征的图像检索与利用。方法/过程] 以Flickr图片为例，利用PMI算法对WordNet-Affect词表进行预处理形成典型情感词表；结合Ekman提出的6类基本情感类型，利用标签对图片情感类型进行标注；并且，通过实验对分类标注效果进行验证；最后，讨论图片特点、标注意图、非情感标签数量对分类标注效果的影响。结果/结论] 研究发现，一幅图片的非情感标签与情感标签在表现图片整体情感类型的倾向性上具有较高一致性；结合PMI算法，利用预处理后的典型情感词表标注图片的结果优于未处理的WordNet-Affect词表；并且，分类标注效果与人工标注结果也具有较好的一致性，其中，快乐类（Happy）和忧伤类（Sad）图片的分类标注一致性最高，惊讶类（Surprise）的分类标注一致性最低；分析发现，仅通过标签标注图片情感类型的过程中，分类标注效果与图片情感的典型性、单一性以及图片发布方和欣赏者意图、动机的差异、图片的非情感标签个数都有关系。
|
| 关 键 词： | 图像 标签 情感标注 PMI |
| 收稿时间： | 2016-06-30 |
| 修稿时间： | 2016-10-20 |
Research on Image Emotional Annotations Based on Social Tags |
| |
| Song Lingchao,Huang Kun.Research on Image Emotional Annotations Based on Social Tags[J].Library and Information Service,2016,60(21):103-112. |
| |
| Authors: | Song Lingchao Huang Kun |
| |
| Institution: | 1. The Library of Nankai University, Tianjin 300350;
2. School of Government in Beijing Normal University, Beijing 100875 |
| |
| Abstract: | Purpose/significance] To index emotion type of images is useful for organizing and searching images by affective clues.Method/process] With the Flickr images, this paper took the algorithm of Pointwise Mutual Information and WordNet-Affect wordlists, dealing with tags to categorize images based on Ekman's six emotion types. Then it conducted experiments to test the results. Finally, it discussed the influence of images' features, indexing motivations and the number of non-affect tags on indexing results.Result/conclusion] Findings indicated for one image, the non-affective tags and affective tags reflected the similar emotion as the whole image.And the typical affective wordlists were also tested useful than before. Moreover, the categorizing results based on typical affective wordlists were accorded with the results categorizing by users. Among 6 emotion types, the majority of Happy and Sad images were categorized correctly. While the Surprise images were categorized worst. Furthermore, the accuracy of categorizing was related to the typicality and unique of emotions of images, the difference between images uploaders and users, and the numbers of non-affective tags in images. |
| |
| Keywords: | image social labeling emotional annotations PMI algorithm |
|
| 点击此处可从《图书情报工作》浏览原始摘要信息 |
| 点击此处可从《图书情报工作》下载免费的PDF全文 |
