Semi-supervised Learning And Its Application In Social Media Analysis

In many real-world learning scenarios, acquiring a large amount of labeled training data is expensive and time-consuming. Semi-supervised learning (SSL) is the machine learning paradigm concerned with utilizing unlabeled data to try to build better classifiers and regression model. In practice, unlabeled data are more accessible than labeled data, so a semi-supervised learning algorithm has a very wide range of applications. With the rapid development of social media, social media produced a large number of unlabeled data, which makes semi-supervised learning is finding wider and wider application of the analysis of the social media. We combined with the characteristics of social media, and apply a semi-supervised learning to the analysis of social media.First, we introduced the social media multimodal data, and introduce the concept of multimodal semi-supervised learning. In social media, there are many multimodal data. For example, in social media image classification problem, we can accord to the image itself contains the visual information of image classification, also can use text information such as image tags and comments to classify. Due to the different modal data to demonstrate the different aspects of the subject, multimodal information is compatible and complementary. Effective use of these multimodal data, we can better grasp the information contained in social media data. We analyzed the related work, combination with the characteristics of social media, and give the definition of the multimodal semi-supervised classification problems. We proposed a multimodal semi-supervised learning algorithm and introduced the steps of the algorithm. We applied it to the classification of social media image.Then, we introduce the method of social media image cropping and a semi-supervised learning algorithm is introduced into the image cropping. This is a kind of automatic image cropping algorithm. Experiments prove that the algorithm can adapt to a variety of rules of composition, and combine the characteristics of social media image, can improve the quality of social media image aesthetics.