Efficient Information Retrieval

1
Efficient Information Retrieval

• Christos Faloutsos
• UMD and CMU

2
Past successes

• IR industrial strength systems
• DBMSs same
• Indexing
• B-trees, hashing for formatted data
• vector model, rel. feedback, LSI etc for text
• R-trees etc for spatial data
• Association rules for data mining

3
Future

• REACH OUT many cross-disciplinary challenges
• SCALE UP massive datasets

4
Specific future steps

• Multimedia
• Data mining
• World-Wide Web
• High performance storage

5
Multimedia

• find similar video clips, images, audio clips
• Machine vision (DWT, edge detection,...)
• voice processing (LPC, HMM, )
• medical doctors (X-ray/MRI db, )
• meteorology, astrophysics,

6
Data mining

• find correlations/rules/outliers
• Collaboration between
• Statistics (SVD, Regression, ARIMA, )
• AI (tree classifiers, ANN, )
• Theory of learning (MDL, Kolmogorov-Chaitin)

height
weight
7

• Cross media learning (find correlations between
  audio and video track)
• forecasting
• topic detection and tracking

8
World-Wide Web

• IRDB
• ranking relevance feedback
• visualization
• MM indexing on the web
• synthesis of information
• summarization
• near-duplicate elimination

9
High Performance storage

• RAID, intelligent disks
• tertiary storage (striping, caching, prefetching)

10
Conclusions

• Cross-disciplinary challenges REACH OUT
• massive datasets SCALE UP