1. Circumventing Data Quality Problems Using Multiple Join Paths Yannis Kotidis, Athens University of Economics and Business Amélie Marian, Rutgers University Divesh Srivastava, AT&T Labs-Research

2. 9/11/2006Amélie Marian - Rutgers University2 Motivating Example Sales TN BAN TN BAN CustName ORN PON Provisioning CustName PON SubPON Inventory PON TN CircuitID Ordering ORNTN TN: Telephone Number ORN: Order Number BAN: Billing Accoung Number PON: Provisoning Order Number SubPON: Related PON What is the Circuit ID associated with a Telephone Number that appears in SALES?

3. 9/11/2006Amélie Marian - Rutgers University3 Motivations  Data applications with overlapping features Data integration Web sources  Data quality issues (duplicate, null, default values, data inconsistencies) Data-entry problems Data integration problems

4. 9/11/2006Amélie Marian - Rutgers University4 Contributions  Multiple Join Path (MJP) framework Quantifies answer quality Takes corroborating evidence into account Agglomerative scoring of answers  Answer computation techniques Designed for MJP scoring methodologies Several output options (top-k, top-few)  Experimental evaluation on real data VIP integration platform Quality of answers Efficiency of our techniques

5. 9/11/2006Amélie Marian - Rutgers University5 Outline  Multiple Join Path Framework Problem Definition  Our Approach Scoring Answers Computing Answers  Experimental Evaluation  Related Work

6. 9/11/2006Amélie Marian - Rutgers University6 Multiple Join Path Framework: Problem Definition  Query of the form: “Given X=a find the value of Y” Examples: Given a telephone number of a customer, find the ID of the circuit to which the telephone line is attached. One answer expected Given a circuit ID, find the name of customers whose telephones are attached to the circuit ID. Possibly several answers

7. 9/11/2006Amélie Marian - Rutgers University7 Schema Graph  Directed acyclic graph  Nodes are field names  Intra-application edge Links fields in the same application  Inter-application edge Links fields across applications All (non-source, non-sink) nodes in schema graph are (possibly approximate) primary or foreign keys of their applications

8. 9/11/2006Amélie Marian - Rutgers University8 Data Graph  Given a specific value of the source node X what are values of the sink node Y?  Considers all join paths from X to Y in the schema graph X (no corresponding SALES.BAN) X X Example: two paths lead to answer c1

9. 9/11/2006Amélie Marian - Rutgers University9 Scoring Answers  Which are the correct values? Unclean data No a priori knowledge  Technique to score data edges What is the probability that the fields associated by the edge is correct  Probabilistic interpretation of data edge scores to score full join paths Edge score aggregation Independent on the length of the path

10. 9/11/2006Amélie Marian - Rutgers University10 Scoring Data Edges  Rely on functional dependencies (we are considering fields that are keys)  Data edge scores model the error in the data  Intra-application edge  Inter-application edge equals 1, unless approximate matching Fields A and B within the same application AB (and symetrically for B -> A) Where b i are the values instantiated from querying the application with value a ABBAand

11. 9/11/2006Amélie Marian - Rutgers University11 Scoring Data Paths  A single data path is scored using a simple sequential composition of its data edges probabilities  Data paths leading to the same answer are scored using parallel composition XabY 0.50.80.6 pathScore=0.5*0.8*0.6=0.24 XabY 0.50.80.6 c pathScore=0.24+0.2-(0.24*0.2) pathScore=0.392 0.40.5 Independence Assumption

12. 9/11/2006Amélie Marian - Rutgers University12 Identifying Answers  Only interested in best answers  Standard top-k techniques do not apply Answer scores can always be increased by new information We keep score range information Return top answers when identified, may not have complete scores  Two return strategies Top-k Top-few (weaker stop condition)

13. 9/11/2006Amélie Marian - Rutgers University13 Computing Answers  Take advantage of early pruning Only interested in best answers  Incremental data graph computation Probes to each applications Cost model is number of probes  Standard graph searching techniques (DFS, BFS) do not take advantage of score information  We propose a technique based on the notion of maximum benefit

14. 9/11/2006Amélie Marian - Rutgers University14 Maximum Benefit  Benefit computation of a path uses two components Known scores of the explored data edges Best way to augment an answer’s scores  Uses residual benefit of unexplored schema edges  Our strategy makes choices that aim at maximizing this benefit metric

15. 9/11/2006Amélie Marian - Rutgers University15 VIP Experimental Platform  Integration platform developed at AT&T  30 legacy systems  Real data  Developed as a platform for resolving disputes between applications that are due to data inconsistencies  Front-end web interface

16. 9/11/2006Amélie Marian - Rutgers University16 VIP Queries  Random sample of 150 user queries.  Analysis shows that queries can be classified according to the number of answers they retrieve: noAnswer(nA): 56 queries anyAnswer(aA): 94 queries  oneLarge(oL): 47 queries  manyLarge(mL): 4 queries  manySmall(mS): 8 queries heavyHitters(hH): 10 queries that returned between 128 and 257 answers per query

17. 9/11/2006Amélie Marian - Rutgers University17 VIP Schema Graph Paths leading to an answer /paths leading to top-1 answer (94 queries) Not considering all paths may lead to missing top-1 answers

18. 9/11/2006Amélie Marian - Rutgers University18 Number of Parallel Paths Contributing to the Top-1 Answer Average of 10 parallel paths per answer, 2.5 significant

19. 9/11/2006Amélie Marian - Rutgers University19 Cost of Execution

20. 9/11/2006Amélie Marian - Rutgers University20 Related Work  Keyword Search in DBMS (BANKS, DBXPlorer, DISCOVER, ObjectRank) Query is set of keywords Top-k query model DB as data graph Do not agglomerate scores  Top-k query evaluation (TA, MPro, Upper) Consider tuples as an entity Wait for exact answer (Except for NRA) Do not agglomerate scores  Probabilistic ranking of DB results Queries not selective, large answer set We take corroborative evidence into account to rank query results

21. 9/11/2006Amélie Marian - Rutgers University21 Conclusion  Multiple Join Path Framework Uses corroborating evidence to identify high quality results Looks at all paths in the schema graph  Scoring mechanism Probabilistic interpretation Takes schema information into account  Techniques to compute answers Take into account agglomerative scoring Top-k and top-few