基于深度学习的多媒体画面情感分析  

Emotional Analysis of Multimedia Pictures Based on Deep Learning

作　　者：刘瑞梅 孟祥增 

机构地区：[1]山东师范大学新闻与传媒学院,山东济南250014

出　　处：《电化教育研究》2018年第1期68-74,共7页E-education Research

基　　金：山东省研究生教育创新计划项目"利用MOOC和微课程开展研究生研究性学习的实践研究"(项目编号:SYDD14102)

摘　　要：多媒体教学软件的画面设计对教学效果无疑有潜在的影响。多媒体画面的色彩、布局、图文搭配等视觉效果对学习者的情感有一定程度的影响。分析多媒体画面对学习者情感的影响,对多媒体课件、网站、微课的设计与制作有积极意义。文章分析了多媒体画面图像的情感特点,建立了12个维度的多媒体画面情感描述模型;设计了9层深度卷积神经网络,利用人工标定的2800幅多媒体画面图像的情感值,对深度卷积神经网络进行训练;利用训练后的深度卷积神经网络、对19525幅多媒体画面的情感进行估计;从媒体类型、教学方式、适用学科、学段四个维度对多媒体画面的情感进行了分析,给出了对多媒体画面设计的情感启示,以期对多媒体教学软件的画面设计提供借鉴和参考。The screen design of multimedia teaching software undoubtedly has a potential impact on teaching effect. The visual effects, such as color, layout and graphic collocation of multimedia pictures have some influence on learners＂ emotions. The analysis of the influence of multimedia picutres on learners＂ emotions has positive significance for the design and production of multimedia courseware, websites and micro-lectures. This paper analyzes the emotional characteristics of the images of multimedia pictures and establishes a 12-dimension emotion description model of multimedia pictures. A 9-layer convolution neural network is constructed, which is trained by the values of 2800 multimedia images calibrated manually. Then, the emotions of 19525 multimedia pictures are estimated by using the trained convolution neural network. The emotions of multimedia pictures are analyzed according to media types, teaching methods, suitable subjects and study sections, and enlightenments are put forward for the design of multimedia pictures for future design of multimedia teaching software.

关 键 词：多媒体画面 情感估计 深度学习 

分 类 号：G434[文化科学—教育学]