Efficient Multi-View Maintenance in the Social Semantic Web Views on Social Networks Query = Subgraph matching query View = Query for which the answer set is maintained as the social network database is updated 1 Multiple Views 2 Merging Views 3 Example Merge 4 Merge Optimality 5 Optimal Merge 6 7 8 9 Matthias Broecheler, Andrea Pugliese, and VS Subrahmanian • On large social networks, multiple views are often maintained concurrently. • Maintaining multiple views is very expensive, in particular for rapidly changing databases. • e.g. Twitter has over 340 million tweets / day IDEA: Very often, view queries have overlapping subgraph structures (bold arcs). If we can overlay the different view queries such that these shared substructures can be matched jointly rather than independently, we can save a lot of time. Social network updates are edge insertions (removals), hence graph merging has to center around the inserted edge type. Developed subgraph matching algorithm that can process merged view queries efficiently. Many possible ways to merge query graphs. We want high connected overlap which results in most savings at update time. Define merged view score as the sum of edge overlaps. Finding the optimal view wrt the merged view score is NP–hard. Our greedy view merging algorithm finds near optimal views in practice. Experiments Compared our merged multi-view maintenance algorithm against standard independent view maintenance. 6 real world social network datasets with up to 540 million edges. Randomly generated 12,000 queries with varying degree of overlap and averaged results over 750 trials. All algorithm implemented in Java on top of the COSI graph database middleware. Performance improvement of the Multi-View Maintenance algorithm on 6 different social networks Applications Maintaining multiple views jointly as a merged view leads to significant improvements. 477% faster than standard view maintenance Applications include: Monitoring social networks Fraud, security applications, alerts Business Analytics Knowledge Discovery Caching frequently asked queries ?v4 ?v3 Health Care ?a1 ?a2 ?v7 ?v6 Business Analytics ?v5 topic topic references references publish tweet associated follows publish expert tweet topic follows ?v13 ?v11 ?v12 topic expert associated tweet references publish topic comments ?v9 ?v8 publish associated Edges mapped by 1 , 2 and 3 Edges mapped only by 2 Edges mapped only by 3 Edges mapped only by 1 LEGEND ?v4 ?v3 Health Care ?a1 ?a2 ?v6 Business Analytics ?v5 topic topic references references publish tweet associated follows publish expert tweet expert topic comments ?v9 ?v11 publish ?v10 publish ?v8 references topic associated Edges mapped by 1 , 2 and 3 Edges mapped only by 2 Edges mapped only by 3 Edges mapped only by 1 LEGEND ?v7 ?v16 ?v14 ?v15 topic topic references publish associated follows publish expert tweet ?v13 ?v12 70.0% 75.0% 80.0% 85.0% 90.0% 95.0% 100.0% 0% 100% 200% 300% 400% 500% 600% 700% 800% 900% Physics Enron Youtube Flickr LiveJournal Orkut Outperforming Improvement Mul2 View Maintenance Performance ?person ?article1 Health Care ?expert ?msg1 ?other ?article2 Business Analytics ?msg2 topic topic references references publish tweet associated follows publish expert tweet Health Care ?expert ?msg topic references associated comments expert ?doc ?person ?author ?article publish publish tweet topic