Federated Graph Management and Querying: Subgraphs, Keywords, and Privacy

Data is crucial for decision-making across various domains, but it faces challenges such as explosive growth, complicated integration, and privacy sensitivity. One major obstacle to collaborative data analytics is the presence of data silos, which are isolated data repositories owned by independent organizations that are incompatible with each other. This prevents collaborative data analytics, which could provide more accurate and enriched results in an
integrated source way. For example, different hospitals may have patient records of similar diseases associated with similar features and treatments. On the other hand, many entities perform actions with others, which are usually modeled as graphs, such as social networks, financial networks, and biomedical networks. This proposal targets the problem of federated graph data analytics, which performs graph query processing over multiple parties of graph data whose owners agree to collaboratively tackle graph data silos under a privacy protection
mechanism.
However, existing federated graph analytics faces two significant challenges. First, some fundamental graph queries are not supported under a federated setting, such as k-core search and subgraph pattern counting, due to the complexity of graph structures. Second, graph privacy protection needs careful design, which brings efficient and effective computation difficulties. Most graph data is centralized, which ensures data privacy and prevents access
outside of data owners. However, data breaches are still possible, and some user data is only stored on their own devices individually, requiring more private query processing to ensure the correctness of query answers. Although centralized solutions cannot address users’ security and privacy concerns completely, they remain the norm.
This collaborative research project aims to explore federated graph data management and querying technologies to enable various collaborative queries and ensure data privacy. The core idea is to develop efficient federated algorithms for subgraph search and counting while enhancing them through differential privacy protection. Our research agenda includes 1) federated graph query processing to support fundamental queries of subgraph search and
counting; 2) federated attributed graph analytics to support practical useful queries of spatialtemporal community search and keyword search; 3) enhancing privacy protection for federated graph analytics under differential privacy, which can be effective against privacy attacks; and 4) developing a federated prototype system by integrating all our developed techniques and collaborating with industry partners. This collaborative research is expected to generate new
techniques and theories for federated graph analytics systems, which can benefit the industry and society in graph databases and analytics.