社交多媒体情感分析及其传播研究将在社会维稳、产品营销、危机公关与信息精准获取等应用中发挥重要作用。融合社交多媒体上下文、用户属性和网络结构特征，研究社交多媒体情感提取方法、传播模式及热点事件情感的传播预测是本项目基本研究思路。基于多领域融合数据，采用半监督学习的朴素贝叶斯和支持向量机模型研究社交多媒体文本情感分析；提取社交多媒体中能影响视听觉情感的视频低层特征，建立V-A（Valence-Arousal）模型提取视频情感信息；采用机器学习方法与复杂网络动力学方法研究社交多媒体及其情感在社交网络上的传播模式；融合社交网络热点事件自身特征、社交用户对事件的参与概率和情感传播模式，实现社交网络热点事件的情感预测。将重点探索解决社交多媒体的上下文信息、低层物理特征与高层情感语义的映射机理，以及社交多媒体情感与其传播模式之间的内在联系这两个具有挑战性的科学问题，为相关应用提供新的理论基础和核心算法。

The study of sentiment and information diffusion pattern analysis on social multimedia plays a crucial role in social stability, product marketing, crisis management and information acquisition. In this project, the key idea is to utilize context information, user property data and network topology in the threefold analysis of social multimedia: methods for extracting sentiment features, sentiment dissemination patterns, and sentiment prediction of hot issues. By using cross-domain datasets, we will investigate the analyzed methods of text sentiment by employing semi-supervised learning methods, such as Naive Bayesian and SVM (Support Vector Machine); explore the extraction approaches of video sentiment by applying the V-A (Valence-Arousal) model; research the sentiment dissemination patterns of social multimedia on social networks by taking machine learning methods and dynamical complex network models; probe into the sentiment prediction models of the dissemination tendency of hot issues on social networks by coupling the characteristics of hot issues, the participation probability of users and the sentiment dissemination patterns. This project will focus on two scientific challenges, which could provide the basic theories and the core algorithms for the relevant innovative applications: the first one is to explore the mapping mechanism between the multimedia’s physical features/context and high-level semantic information; and another one is to investigate the latent relation between social multimedia sentiment and the dissemination patterns.