Visual information is critical in many applications, and the retrieval of images is facilitated via keyword descriptors of the image contents. However, the semantic gap presents an arduous task for content-based image retrieval (CBIR) investigations, particularly with abstract and location vocabulary types. In this study, we deployed the k-NN and AdaBoost learning algorithms to compare classification performance between concrete, abstract, and location types in keyword categorization. With a large vocabulary classification (190 concrete, 138 abstract, and 119 location classes from the Corel image collection), AdaBoost rendered the most assignable keywords and achieved a significant improvement in accuracy measures, making it an effective classifier in the one-versus-the-rest mode of operation.