Recent advances in computing infrastructure constitute edge nodes prominently (e.g., smartphones, cars) in their computation pipeline. Applications that combine sensor/IoT data from edge devices in a distributed fashion are growing. Dynamically composed applications that combine resources from edge nodes and the cloud are becoming common in various domains including urban and military settings. A key challenge in such applications is to bridge the gap between the application's description and its IoT resource requirements, where the application description is unstructured text and its IoT requirements are structured. In this paper, we describe an approach that develops a model which given unstructured text description of the service predicts its IoT sensor requirements. Our model can predict with an average accuracy of 77% and up to 88% on the top-20 sensor requirements for 300 different applications.