Research On Deep Learning And Its Application In Social Media Analysis

As the rapid development of the Internet, there exist a large volume of social media data such as images, videos and text. In addition, some new social media focusing on ima ges like Flickr and Instagram become more and more popular as the result of the develop ment of mobile devices which make it more convenient to take and edit pictures. It’s bee n a hot-debated research area to effectively manipulate that large amount of image data a nd perform data mining in order to improve user experience and enhance marketing strate gies for enterprise. Since 2006, deep learning has emerged as a new area of machine learning research. Human information processing mechanisms (e.g., vision and audition), however, suggest the need of deep architectures for extracting complex structure and building internal representation from rich sensory inputs. It is natural to believe that the state-of-the-art can be advanced in processing these types of natural signals such as images in the social media if efficient and effective deep learning algorithms can be developed. To cure the above problems, this paper presents a classification algorithm of social media image semantic based on deep learning model.This thesis introduced the essentials of Deep Learning algorithm, training algorithm and their main differences between neural networks. Several broadly used Deep Learning models are also elaborated in this thesis. Secondly, applications and available methods of Deep Learning in image semantic classification problem are described; image semantic classification models based on Stacked denoising Auto-Encoders (SdAE) and Convolution Deep Boltzmann Machine (CDBM) are also presented in the thesis. Experimental proof is perform to demonstrate that the two models are effective in the image semantic classification. Thirdly, the thesis described the application and available methods of Deep learning in image aesthetic quality assessment and the image aesthetic quality assessment framework based on the Convolution Neural Network and Support Vector Machine. In addition, this thesis proposes a conclusion for the whole work, which may serve for future research and study.