Building Concept Hierarchies for the Internet of Things Patterns Using Domain-specific Dependency Knowledge

Abstract

The number of the Internet of Things (IoT) patterns in the literature is growing rapidly. As a result, this makes it difficult to identify and differentiate a pattern from a large number of related patterns. Thus, there is a need for organizing these patterns in a meaningful way to facilitate speedy retrieval and guide the IoT architects in building a classification scheme that groups a family o f r elated patterns together. The classification of these patterns is often made difficult by the fact that the names of these patterns generally do not reveal the core essence of the pattern. In order to understand the essence of a pattern, users are generally expected to go through several pages which may still be obscure and difficult to understand due to semantic barriers and richness of language. Intuitively, this problem can be addressed by assigning a few verbal predicates that best describe the core essence of each pattern. In this paper, we show that Formal Concept Analysis (FCA) and Concept Lattices are suitable tools to support this task. Accordingly, we make use of FCA to build a concept lattice, which serves as a semantic index to model terms that define the core attributes of each pattern. We introduce the notion of attributes hierarchies to scientifically identify the one main concept that seems to underlie the meaning of each IoT pattern. The more significant attributes for the pattern are represented by concepts that branch out of the root node concept, forming leaf nodes down the hierarchy. This concept lattice feeds from information taken from a few pre-identified sentences taken from a document. These are sentences that describe the core attributes of the pattern. By quantifying sentence similarity between these preidentified sentences and other sentences in the document, we can identify sentences from which we can extract concepts for building the concept lattice. Experimental results show a promising performance in using this approach for organizing the IoT patterns.