1 / 39

Probabilistic Ranking of Database Query Results

Probabilistic Ranking of Database Query Results. Surajit Chaudhuri , Microsoft Research Gautam Das, Microsoft Research Vagelis Hristidis , Florida International University Gerhard Weikum , MPI Informatik Presented by: Ranjan alankar raju Sindhu satyanarayana. AGENDA.

lanza
Download Presentation

Probabilistic Ranking of Database Query Results

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Probabilistic Ranking of Database Query Results SurajitChaudhuri, Microsoft Research Gautam Das, Microsoft Research VagelisHristidis, Florida International University Gerhard Weikum, MPI Informatik Presented by: Ranjanalankarraju Sindhusatyanarayana

  2. AGENDA • Introduction & Motivation • Problem Definition & Architecture • Definition of Ranking Function • Implementation • Experiments • Conclusions & Limitations

  3. LET US SEE THE • Introduction & Motivation • Problem Definition & Architecture • Definition of Ranking Function • Implementation • Experiments • Conclusions & Limitations

  4. Introduction and Motivation REALTOR_DB

  5. PROBLEM DEFINITION- MANY ANSWERS • SELECT * FROM REALTOR_DB WHERE CITY=‘SEATTLE’ ; RESULT OF THIS QUERY: Too Many Answers

  6. PROPOSED SOLUTIONS • QUERY REFORMULATION TECHNIQUES: -BY PROMPTING THE USER • AUTOMATIC RANKING: -USING GLOBAL AND CONDITIONAL SCORE

  7. LET US SEE THE • Introduction & Motivation • Problem Definition & Architecture • Definition of Ranking Function • Implementation • Experiments • Conclusions & Limitations

  8. DEFINITIONS AND SYMBOLS • What are Specified Attributes (Denoted as ‘X’) • City • What are Unspecified Attributes (Denoted as ‘Y’) • View • Price • SchoolDistrict • BoatDock

  9. PROPOSED RANKING FUNCTION • Global Score : Global importance of unspecified attributes Eg: VIEW=‘WATERFRONT’ • Conditional Score: Correlations between specified and unspecified attributes Eg: If CITY=‘SEATTLE’ and VIEW=‘WATERFRONT’ Will BOATDOCK=‘YES’ interest him?

  10. ARCHITECTURE

  11. RANKING FUNCTIONSRules & Theorems For PIR • Bayes’ Rule: p(a/b) = [ p(b/a) p(a) ] / [p(b)] Product Rule: p(a,b/c) = p(a/c) * p(b/a,c)

  12. BAYES’ THEOREM EXAMPLE • 1% of the population has X disease.. A screening test accurately detects the disease for 90% of people with it. The test also indicates the disease for 15% of the people without it ( the false positives). Suppose a person screened for the disease tests positive. What is the probability they have it?

  13. BAYES’ THEOREM Cont… • Interpretation and Assumption: D - Event that person has disease T- Test is Positive • Given: p(D)= 1% p(D|T)=? p(T|D) = 90 % p(T|D’)=15%

  14. Tree structure Interpretation Four Cases 1. (D n T)-Has disease and test +ve. 3. (D’ n T)- No disease and test +ve. 2. (D n T’)-Has disease and test –ve. 4. (D’ n T’)- No disease and test –ve. 1 D’ D T T T’ T’

  15. LET US SEE THE • Introduction & Motivation • Problem Definition & Architecture • Definition of Ranking Function • Implementation • Experiments • Conclusions & Limitations

  16. Rules & Theorems For PIR cont… t-Tuple (Document) R-Relevant Documents R- Irrelevant Documents

  17. Adaptation of PIR • Partition tuple ‘t’ into two parts t(X) and t(Y) • Replacing t with ‘X’ & ‘Y’

  18. Adaptation of PIR cont… • QUERY SPECIFIED BY USER: Select * From Realtor_db where City=‘Seattle’ and Price=‘High’; • FINAL RANKING: • Waterfront Views • Greenbelt Views • Street Views

  19. Limited Independence Assumption • X (and Y) values within themselves are assumed to be independent. • Dependencies between the X and Y values are allowed

  20. Eliminating R Incoming Query: Select * from Realtor_db where City=‘Seattle’;

  21. Workload-Based Estimation FINAL RANKING FORMULA Where: p(y|W) = Relative frequency of unspecified attribute ‘y’ given workload ‘W’ p(y|D)= Relative frequency of unspecified attribute ‘y’ given data base ‘D’ p(x|y,W)=Frequency of correlation between x and y in W P(x|y,D)=Frequency of correlation between x and y in D

  22. Detailed Process

  23. LET US SEE THE • Introduction & Motivation • Problem Definition & Architecture • Definition of Ranking Function • Implementation • Experiments • Conclusions & Limitations

  24. IMPLEMENTATION • Preprocessing: 1. Computation of modules: p(y | W), p(y | D), p(x | y, W), and p(x | y, D) for all distinct values of x and y. 2. Storing these atomic probabilities as database tables in intermediate knowledge representation layer with appropriate indexes. 3.Computation of index module resulting in conditional and global lists table.

  25. IMPLEMENTATION cont… CONDITIONAL LISTS Cx: Contains <TID, CondScore> in descending order GLOBAL LISTS Gx: Contains <TID,GlobScore> in descending order

  26. IMPLEMENTATION cont…

  27. Conditional and Global Scores

  28. Conditional and Global List tables

  29. IMPLEMENTATION cont… • Query Processing Component.

  30. List Merge Algorithm contd...

  31. LET US SEE THE • Introduction & Motivation • Problem Definition & Architecture • Definition of Ranking Function • Implementation • Experiments • Conclusions & Limitations

  32. EXPERIMENTS • Datasets: • MSN HomeAdvisor database • Internet Movie Database(IMDB)

  33. Quality Experiments • Examples of Ranking Results: Query: select * from SeattleHomes where City=‘Seattle’ and Bedroom=1; • Conditional ranked condos with garages the highest • Global failed to recognize importance of the unspecified attribute Garage=‘Y’

  34. Quality Experiments • User Preference of Rankings: • Users given top 5 results of rankings for 5 queries • Ranking preferred by users indicated below:

  35. LET US SEE THE • Introduction & Motivation • Problem Definition & Architecture • Definition of Ranking Function • Implementation • Experiments • Conclusions & Limitations

  36. CONCLUSION & LIMITATION CONCLUSION: Automated approach leverages data and workload statistics and correlations. LIMITATION: Existence of correlations between text and non-text data.

More Related