The Semantic Web marks the next cornerstone in the development of the World Wide Web, and constitutes the ultimate goal of enabling machines to access and interpret data published on the Web automatically. This, in turn, allows applications to grasp the meaning inherent in the data without human assistance, and empowers them to reason about it autonomously. In practice, however, the task of reasoning about Semantic Web data comes with a number of issues attached. Among these are typical problems related to databases, like incomplete information or conflicting data, but also new challenges came up, like reasoning at the scale of the World Wide Web or the effective construction of expressive knowledge bases. It turns out, though, that many of the obstacles encountered are actually related to the methods employed for reasoning with classical logic rather than to the nature of the Semantic Web itself, and so people started to investigate machine learning as an alternative approach to reason about Semantic Web data This thesis follows that very path and investigates the use of deep learning in order to tackle the problem of instance checking, a particular kind of reasoning task commonly encountered in real-world applications of the Semantic Web. In doing so, I develop a new kind of deep model, and compare its performance on this task with state-of-the-art approaches based on machine learning. Thereby, I show that the newly introduced model is indeed competitive and in parts even superior to existing state-of-the-art techniques. However, the main contribution of this thesis is a generalization of recursive neural tensor networks to arbitrary relational datasets, with the single restriction that these data involve binary relations only. In contrast to similar approaches, a distinctive property of this model is that it can be applied to relational datasets out-of-the-box, i.e. we do not have to account for the particular structure of a dataset beforehand. Deep learning is among the hottest topics within the field of machine learning right now, and has - to the best of my knowledge - not been employed in connection with the Semantic Web so far.