finding all employees who had been in
the meeting rooms that reported an abnormal energy consumption. However,
we have not discovered Io TSE prototypes
taking advantage of these correlations.
Therefore, for the sake of clarity, we will
not depict detailed thing-thing correlations in the following discussions.
Figure 4 depicts the IoT infrastructure in a smart building as a heterogeneous graph (upper) and the derived
meta-graph. IoT things in this illustration consist of a smart light bulb, a
meeting room, and a staff member who
uses these facilities. The light bulb has
seven Io T content items of four classes.
The meta-graph captures relationships
between Io T content types and things,
as well as among things. For instance,
two representative content items of
the lightbulb are captured in the meta-graph as a single link between the representative content type and a thing.
A meta-path is a sequence of edges
on the meta-graph from one type of Io T
content, through various Io T things, to
another type of Io T content. In the Io TSE
context, each meta-path can model the
relationship between a type of Io T content used for assessing query and a type
of Io T content used for deriving search
results. By aggregating multiple meta-paths, we can model an Io TSE instances
that utilize multiple types of Io T content.
A meta-path can be represented as
(Query content type)→ Things →
(Result content type).
To demonstrate the meta-path model, we will model an Io TSE instance that
queries for “homepages of Io T-enabled
light bulbs which are reporting an abnormal energy consumption” as an example. This query can be decomposed
into two subqueries: “finding the virtual representative of things, which are
lightbulbs” and “finding the virtual representative of things, which are reporting an abnormal energy consumption.”
The first subquery involves assessing
each discovered representative (
Representative) to determine whether it belongs to a lightbulb (Thing) and returning the representative of that lightbulb
(Representative) as the search result.
This subquery can be modeled with the
The second subquery involves as-
content types to assess a query and to
derive search results. Moreover, the
types of IoT content appearing in a
query influence the internal operations
of an Io TSE instance. 20 As a result, an
IoTSE model must capture succinctly
both the types of involving Io T content
and the relationships among those
types. Terms such as “object search”
are inadequate. To address this issue,
we propose a model called meta-path.
Before defining meta-path, it would
be helpful to introduce the idea of mod-
eling the Internet of Things as a het-
erogeneous graph, which was inspired
by PathSim. 17 The nodes in this graph
consist of Io T contents and Io T things
that own these contents. The edges
that link things and content denote a
possessive relationship between them.
The edges that link things denote their
possible correlations, such as sharing
owners or operation environments. 23, 24
From a concrete graph, we can de-
rive a meta-graph that presents rela-
tionships between types of nodes. Each
node in a meta-graph is either a type of
Io T content or a thing. An edge between
a content type and a thing represents
the content type is offered by the thing.
An edge between two things represents
a correlation between them. Different
types of thing-thing edges represent
different forms of correlation between
things. These thing-thing relationships
can enable interesting queries such as
Figure 3. Four types of Io T content of an Io T-enabled lightbulb.
Figure 4. (Upper) Meta-graph from a concrete Io T network, and meta-path R + D →T→ D.
R: JSON description
R:H TML description
S: Energy consumption
A: “Power on” service
A: “Power off” service
A: “Change light colour” service