IEEE Systems, Man and Cybernetics Magazine - October 2022 - 49

influence of various neighboring points on the entity. To
deal with the data-sparse problem, the multihop relationship
information among entities is calculated at the same
time. Finally, a decoder is used to collect global information
among dimensions. Link prediction experiments
show that the multisource information
combined knowledge representation
learning (XAI-CNN)
model based on explainable AI
(XAI) can effectively use multisource
social IoT information
beyond triples and that other techniques
may be better than the
baseline model.
Knowledge Representation
Learning Models
In recent years, a variety of knowledge
representation learning models
has been proposed. Inspired
by word2vec [2], TransE [1] regards
the relationship in the triplet as the translation among
the head entity and the tail entity and obtains a large
performance improvement with fewer parameters, but it
is suitable only for dealing with one-to-one relationships,
so many extended models are derived from it.
TransH [1] introduces a specific relation hyperplane,
which uses the hyperplane normal vector and translation
vector to represent the relation vector, and maps
entities to the hyperplane space of different relations.
TransR [3] further assumes that relations should have
their own semantic space and that entities should be
regarded as vectors in the semantic space. Therefore, it
defines transition matrices for different relations, mapping
entities into the corresponding spaces. RESCAL [4]
uses vectors to represent entities to capture the implicit
semantic information and uses matrices to represent
relationships to simulate the pairwise interaction
among various elements. DistMult [5] greatly reduces
the parameters in the RESCAL model by restricting the
relational matrix to be diagonal, but it is suitable only
for simulating symmetric relations.
HolE [6] combines the advantages of RESCAL and DistThe
encoder receives
the entity vector
acquired from the
encoder, which is
used to continue
training, and the
outcome is the
triplet's final score.
learning (XAI-CNN) model. XAI-CNN includes a TransE
model-based encoder that combines the structural triples
in the knowledge graph with the textual description information
of the entity, the hierarchical-type information of
the entity, and the structural information of the graph to
learn the information described by
complex relationships. The hierarchical-type
information of entities
can assist individuals in automatically
connecting distinct entities
and constraining the semantic
properties of the entities through
the type of information to which
the entities belong.
The entity's description information
is a more complete explanation
of the entity's relevant content,
which contains a lot of critical
knowledge and is a valuable addition
to the triple structure. The
graph's topological information
explains the connections among distinct items and may
accurately depict their spatial interactions. By combining
them with the original triples, the knowledge graph's
hidden entity and relation properties may be better captured.
XAI-CNN, on the other hand, employs the ConvKB
model as a decoder to obtain the global features of triplet
vectors in multiple dimensions while keeping the
model's translation characteristics. The encoder receives
the entity vector acquired from the encoder, which is
used to continue training, and the outcome is the triplet's
final score.
hd
td
hs
ts
Hierarchical-Type
Projection
Hierarchical-Type
Projection
r
Encoder
Mult. First, the head and tail entities in the triplet are combined
through cyclic correlation operations and then
semantically matched with the relation vector, which not
only has the powerful representation effect of RESCAL but
also has DistMult simplicity. To fully simulate asymmetric
relations, ComplEx [7] applied complex numbers to knowledge
representation learning for the first time to solve the
asymmetric problem in DistMult. The interpretability
problem is a technical term for the difficulty of explaining
AI choices.
The XAI-CNN model, which is a typical encoder-
decoder structure, is proposed in this study as a multisource
information combined knowledge representation
Graph Attention
Networks
+
Graph Attention
Networks
Decoder
ConvKB
Figure 1. The overall framework of XAI-CNN for the
social IoT.
October 2022 IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE 49
IEEE Systems, Man and Cybernetics Magazine - October 2022

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - October 2022
