Synergies Between Graph Data Management and Machine Learning in Graph Data Pipeline.

Graphs are popular mathematical tools to model data with relations, such as the Web, social and biological networks, financial transactions, and knowledge bases. Machine learning and recently, deep learning over graphs becomes preva-lent. In modern data science applications, complex data move through various processes involved in machine learning to gen-erate the final predictive output, thereby creating a data pipeline consisting of graph data extraction, acquisition, and cleaning, graph embedding, machine learning training and inference, downstream tasks, explainability, and adding human in-the-loop, as depicted in Figure 1. We investigate how graph data management, which deals with effective, efficient, scalable, and user-friendly systems and algorithms for storing, processing, and analyzing large volumes of heterogeneous and complex graphs, could benefit from graph machine learning and vice versa, over the end-to-end graph data pipeline. We shall emphasize on (1) how graph data management helps in graph machine learning, e.g., in scalable graph embedding and designing user-friendly explainability methods; and (2) how graph machine learning helps in graph data management, e.g., in question answering over knowledge graphs.