Things search engines play a key role in increasing the visibility of the emerging Internet of Things (IoT) paradigm. Developing an innovative search approach is a fundamental step to lay the foundations of future IoT search engines. Currently, the most adopted approach for searching things is based on keyword search. Unfortunately, keyword search does not provide enough functionality for an IoT search engine. Correlating things based on their attributes is an emerging approach which can potentially improve the IoT search process. Since in reality there might exist a number of different correlations between a pair of everyday objects, integrating and applying them in IoT search is challenging. In this paper, we propose the ECS (Extract, Cluster, Select) framework. Our framework contains a novel approach to extract and integrate different types of correlation graphs with a spectral clustering method and a selection method to improve the coherence and the diversity of top-k results for a given search query. We evaluate our framework through extensive experiments using real-world datasets from different domains of IoT applications. The results show that the quality of search results improves greatly after we diversify the results of IoT data queries.