Internet of things (IoT) is bringing Internet truly into our routine life by deploying intelligent equipments ranging from multi-modal sensors to white intelligent goods [1]. As Cisco predicts in [2], over 50 billion devices will be joined into the Internet before 2020. Five hundred zettabytes of data will be produced by tremendous machines, devices, and even the interactions between them. Moreover, the developments of IoT give birth to intelligent realms such as smart transportation [3], e-health [4] and smart homes [5] which are aimed at providing users with better service and higher quality of life. However, due to the lack of interoperability, information generated by different sensors or devices cannot be shared with each other, which has become a severe challenge nowadays. Although nearly 45% data created on the Internet can be processed, it is tough work to mine and dig out the hidden information behind them. Moreover, cross-domain knowledge becomes increasingly difficult to share with others because of the heterogeneity of data. To achieve a better interpretation of heterogenous data, more and more researches start to focus on techniques enabling machines to intelligently understand IoT data. Among all approaches, adding semantics to IoT data is one of the most prevalent methods. Known as an extension of the World Wide Web [6], the semantic web resolves isolation problems between heterogenous information and provides a better understanding of surroundings. By adding general mark-ups and notifications, semantization makes it possible for machines to understand and interpret heterogenous data and prompts cross-domain interactions to a large extent. This paper illustrates an overview of IoT data semantization, including related concepts, architectures, key techniques, applications and challenges. The main contributions of this survey are as follows:

• It provides a detailed overview of data semantization such as the related concepts and existing architectures for adding semantics to IoT data and summarizes a general processing architecture for data semantization.

• It presents key techniques involved in data semantization including techniques in data collection, data preprocessing and semantic annotation.

• It analyzes challenges and open issues that are worth studying in future work such as standardization and generalization, complexity and dynamicity, and security and privacy.