Knowledge extraction in cyber-physical systems (CPS) is a challenging task. Traditional methods are not efficient for this complicated task; an approach that exploits parallel and distributed computing methods while considering scalability is essential. This paper proposes a distributed semantic network (DSN) architecture that can be fitted to the well-known MapReduce platform, utilizing its parallel and distributed processing capability to arrange and manage semantic knowledge base construction and management.

The introduction and literature review explain the traditional methods of data representation, vector-based representation, graph-based representation, and bag-of-words. Knowledge extractions from semi-structured and unstructured data are detailed. Multiple order semantic parsing (MOSP) is counted as the basis of heterogeneous data analysis, and reasons for such an approach to distributed knowledge base extraction and construction are expressed.

A distributed semantic network is the key topic of the paper. It is constructed based on a multi-layer hierarchical graph. Using MOSP, following subject-predicate-object and generalization-specialization between the terms in the text, a graph is created. A multiple order semantic tree is the foundation of this structure. It is constructed by splitting the statement into semantical parts, like word, phrase, and symbol, via multi-order semantics. The order is refined to represent the hypernym-hyponym relationship in the tree layers and the subject-object relationship at its nodes and binary branches. Also, regarding scalability, semantic expansion is the main topic.

An important objective relates to feeding the proposed architecture into a parallel and distributed platform for execution. Illustrating the matter, the paper presents “scalable DSN knowledge management,” and then knowledge extraction and knowledge refinement are discussed. In the MapReduce utilization section, “with the help of scalable DSN,” which so far is transparent, utilization of the MapReduce platform is supposed to be fulfilled.

The introduction to hierarchical DSN and an implementation of its lemma are a good start, but the rest of the paper can be confusing. Despite the paper’s frequent emphasis on the platform’s scalability, it fails to explain the scalability mechanisms. Moreover, there is not even one statement about the relevance of the proposed architecture to CPS, as the paper claims.