Abstract:The essence of connection in vehicle network is the social relationship between people, and thus Vehicular Social Networks (VSNs), characterized by social aspects and features, can be formed. The information collected by VSNs can be used for context prediction of autonomous vehicles. Multivariate relations are common in square connected relations caused by geographic characteristics in VSNs. They can effectively reflect the high-order structural features of the network dataset. It is necessary to exploit the multivariate relations of VSNs to improve the performance of context prediction. However, The representation of entity-relationes in the network often adopts a binary form, and the existing graph learning methods rely on the neighborhood information of nodes to achieve the aggregation or diffusion of information. Using this to represent multivariate relations will result in partial omissions or even complete loss of valuable information, which ultimately affects the learning effect of learning methods. In order to better understand the social behavior of the VSNs, this paper uses the network motif to implement the representation of the multivariate relations in the network, and proposes the graPh learnIng with moTif mAtrix (PITA) method. This method can be used as a preprocessing step for the measurement strategy of the relations in VSNs and the graph learning, which can mine the information in VSNs and improve the accuracy of the original graph learning method by the multivariate relation information. We performed experiments on 6 network datasets. The experimental results show that in the node classification task, the baseline method modified by the PITA method has a higher classification accuracy than the original method.

Multivariate Prediction for Learning in Relational Graphs

Multivariate Prediction for Learning on the Semantic Web

Link Prediction in Multi-relational Graphs using Additive Models.

Predictive Modeling using Features derived from Paths in Relational Graphs

Graphical Models for Relations - Modeling Relational Context

Scalable Relation Prediction Exploiting Both Intrarelational Correlation and Contextual Information.

Multivariate Relations Aggregation Learning in Social Networks

A Semantic Matching Energy Function for Learning with Multi-relational Data

Multi-task Neural Network for Non-discrete Attribute Prediction in Knowledge Graphs

A Probit Tensor Factorization Model For Relational Learning

Link Prediction in Multi-Relational Networks Based on Relational Similarity

A Review of Relational Machine Learning for Knowledge Graphs

Application of Graph Learning With Multivariate Relational Representation Matrix in Vehicular Social Networks

Link Prediction with Relational Hypergraphs

Supervised Learning on Relational Databases with Graph Neural Networks

Generalized Relation Learning with Semantic Correlation Awareness for Link Prediction

Relational Knowledge Prediction Via Dynamic Bi-Mode Embedding.

Learning Word Representations from Relational Graphs

Efficient Relational Learning with Hidden Variable Detection.

Topic-Based Embeddings for Learning from Large Knowledge Graphs

Learning Class-Level Bayes Nets for Relational Data