Research at

1

• Research at
• Open Systems Lab
• IIIT Bangalore
• http//osl.iiitb.ac.in/

2
Broad Areas

• Co-occurrence analyses
• Multi-agent approaches for (database related)
  optimization

3
Co-occurrence Analyses

• Using models of semantic memory from cognitive
  psychology to extract latent semantics in
  document collections

Graphs depicting higher-order inferences
Co-occurrence (labeled) graph
Document corpus
4
Co-occurrence graph

• Captures pair-wise co-occurrences across
  different typed entities
• Entity types
• Nouns (Person, Institution, Place, Country, etc.)
• Tags
• URLs
• Phrases

5
Higher-order inferences

• Topic anchors
• Topic markers
• Synonymy
• Semantic siblings
• Topic induction

6
Higher-order inferences

• Co-citations as URL co-occurrences
• Contrasting co-citation patterns between Web
  pages and Wikipedia
• Co-citation as hyperlink endorsements
• Co-citation as knowledge aggregation
• Co-citation as conditional probability of topical
  relevance

7
Dataset

• Co-occurrence graph built from a complete
  Wikipedia dump
• Co-citation graph built from a crawl of over 10
  million pages and over 85 million hyperlinks

8
Some results

• Topical anchor experiments
• http//tinyurl.com/topicalanchors

9
Some results web co-citation graph
10
Some results endorsed hyperlink graph
11
Some questions

• Innate macro characteristics of co-occurrence
  graphs
• Concept formation from instances of
  co-occurrences
• Multipartite clustering

12
Multi-agent optimization

• Query optimization in stream grids
• Distributed index design under arbitrary
  constraints (churn, load, symmetry, etc.)

13
Thank you