Hyperbit: A Financial Temporal Knowledge Graph Data Storage System

Relationships or interactions among entities interactions often have occurrence time. So, temporal graph is becoming a popular model to represent temporal data. Temporal graph is generally much larger than corresponding non-temporal graph because an non-temporal edge may have many corresponding temporal edges. It raises challenges for querying temporal graphs. Here, we present HyperBit, a temporal graph store which can answer temporal queries efficiently. HyperBit models temporal labeled graph as a series of updates or log records on graph. Then we design an efficient partition storage for log. Since it is costly to answer temporal queries using logs due to full scan of log, we propose an optimal algorithm to build some snapshots to speedup query processing. So, HyperBit can answer temporal queries by applying log records on the snapshot close to the time in query. HyperBit employs SPARQL instead of a new language to describe temporal queries. Thus, HyperBit can process non-temporal queries on temporal/non-temporal graphs. Extensive experiments show that HyperBit significantly outperforms RDF-3x, Jena-TDB in terms of update speed while it has a compact storage. When querying static graphs, HyperBit also outperforms RDF-3X, Jena-TDB by a wide margin and is on par with TripleBit. For temporal queries, HyperBit can easily handle billion graphs, maintaining linear time growth so that has excellent scalability.