A Hybrid Semi-supervised Topic Model

Latent topic models are used to analyze the low-dimensional semantic meaning of documents and images, which are widely applied to object categorization. However, object labeling is expensive and subjective in real applications. Thus, a hybrid semi-supervised topic model is proposed, which uses a small amount of labels to help the generative topic model find semantic topics and cluster the unlabeled data to the same class. We applied the model to obtain the semi-supervised LDA and pLSA methods. Experimental results on natural scene and head pose classification tasks show that the proposed method remains promising using only partial labels in the training process, which demonstrates the effectiveness of the proposed method.