Experts in big graph analysis systems met up at DIKU
The workshop in Big Graph Analysis Systems on 21-22 August 2017 collected renowned international researchers as well as local Danish industry experts, who brought a variety of perspectives and depth of work for discussion in Copenhagen.
The purpose of the workshop was to get together experts from several relevant communities that are actively addressing the problems related to big graph data analysis to identify commonalities and synergies and to stimulate cross-collaboration that can be expected but yet to be explored. The workshop featured invited talks by world-renowned researchers and extensive time for discussions.
Themes of discussions in the workshop
For concreteness, we highlight a few themes of discussion among many in the workshop. The modeling, querying, and storage of graphs currently receive attention from many researchers, with particular emphasis on the relationship between graphs and relational databases. Perspectives included the modeling of graphs as views over relational data (Deshpande), querying and summarization of partially implicit RDF graphs reusing relational database engines (Manolescu), benchmarking and query generation for evaluation of systems supporting graphs (Fletcher), compilation of conjunctive relational queries into factorized graphs (Olteanu), graph programming models and efficient implementation that can be integrated with relational databases (Chafi), as well as storage and indexing techniques for graph queries over relations (Papakonstantinou).
Another theme of interest is in the processing and support for machine learning and analytics over graph data. Researchers showed strong interest in developing solutions for improving such type of analytics in processing speed or prediction quality. Perspectives included the generalization of incremental view maintenance in databases through task-specific rings, e.g., for high-performance linear regression over streaming data (Nikolic), as well as investigation of deep learning techniques for better prediction in chemical bioactivity data (Huan). Local industry experts brought about two different additional examples of the need for this type of analytics, one in the context of reverse engineering of a social network for the web advertisement market (Kefaloukos) and one in the context of entity reconciliation in the information spaces of enterprise systems as varied as CRM, email, social, and web solutions (Ward).
Furthermore, the workshop collected ideas on emerging graph workloads and system architectures. Among these, we can enumerate the challenge of implicit graph search (Düdder), the survey and analysis of differences in graph workloads and processing between industry and academic research (Salihoglu), and the perspectives on the use of GPUs as a platform for graph analysis (Zhou)