SQL or NoSQL? Contrasting Approaches to the Storage, Manipulation and Analysis of Spatio-temporal Online Social Network Data

摘要

Researchers are now accessing millions of Online Social Network (OSN) interactions. These are available at no or low cost through Application Programming Interfaces (APIs) or data custodians including DataSift and GNIP. Records held in Extensible Markup Language (XML) or JavaScript Object Notation (JSON) are well structured but often inconveniently formatted for use in popular Relational Database Management Systems (RDBMS) or Geographic Information Systems (GIS) software. In contrast, emerging NoSQL (Not-only Structured Query Language) technologies are specially designed to 'ingest' unstructured data. Extract/Transform/Load (ETL) procedures for the storage and subsequent analysis of two OSN datasets in SQL/NoSQL databases are examined. The fixed data model of the relational approach may prove problematic when loading unpredictable document-based structures arising from extended periods of data collection. Although relational databases are far from obsolete the spatial analysis community seems likely to benefit from experimentation with new software explicitly designed for handling spatio-temporal Big Data.