Semantic Representation With Heterogeneous Information Network Using Matrix Factorization for Clustering in the Internet of Things.

Emerging communications technologies, such as IPv6 and 5G, will enable massive numbers of devices to connect to the Internet of Things. With the scale of networking equipment increasing, how to effectively use extensive IoT data is an increasingly urgent issue. The interaction relationships between IoT devices based on various application requirements have not received enough attention in traditional data mining methods. However, IoT devices require a large amount of information to interact in practical applications every moment, which produces a variety of semantic relationship. Different from the traditional methods, this paper employs graph structure to represent the semantic relations of IoT data, which exploits fine-grained semantic information more efficiently. Our contributions are as follows: 1) we propose the IoT data representation framework by converting semantic relations into graph structure, 2) the framework can leverage different meta-paths in the graph to measure the similarity among entities in the IoT from different perspectives, and 3) we conduct a cluster experiment based on regularization improved matrix factorization for different application scenarios that consider the semantic similarity. We demonstrate our method using a real-world dataset, and the experimental results show the practicability and effectiveness of our proposed approach. This paper presents a new research angle to analyze semantic data in the IoT.