ESTEEM: Trust-aware P2P data integration Carola Aiello,Tiziana Catarci, Diego Milano, Monica Scannapieco Dipartimento di Informatica e Sistemistica Università di Roma “La Sapienza” 1 Outline Progetti precedenti Obiettivi ESTEEM Problematiche e direzioni di ricerca dell’unità Data quality: Quality-aware query processing Privacy: Privacy-aware record matching Trust: Modello di trust per le sorgenti 2 DaQuinCIS project (2003) MIUR – COFIN/PRIN Main focus: data quality in cooperative information systems (CISs) Data Quality Problems: Record Matching Quality-driven query processing 3 Motivations A real example: e-Goverment project to integrate data about Italian companies Query Company XYZ ? DATA INTEGRATION LAYER Chambers of Commerce Social Insurance Agency Accident Insurance Agency4 Id Chambers of Commerce Name Type of activity City Address Social Insurance Agency Accident Insurance Agency 5 The Three Real Records ID Type of Activity City CNCBTB765SDV Retail of bovine and ovine meats 0111232223 Grocer’s shop, Which beverages CNCBTR765LDV Novi Ligure Name Meat production of Bartoletti Benito Pizzolo Bartoletti Formigaro Benito is the actual company XYZ to Meat ? be returned to the client production Butcher • One of 3 ? Which ? • A “merge” of the 3 ? Ovada Meat production in Piemonte of Bartoletti Benito Address National Street dei Giovi 9, Rome Street 4, Mazzini Square 6 Objectives of the Research 1. Given a set of distributed and heterogeneous data sources that are affected by data quality problems Improving the quality of each data source 2. Record matching across sources Provide a unified and trasparent access to data sources Data Integration & Quality-driven query processing 7 Improving quality of addresses in Italian PA (2004) Accordo di collaborazione AIPA (ora CNIPA) e ISTAT Aprile 2002-Luglio 2004 Proposta di formati standard per l’acquisizione e l’interscambio degli indirizzi Proposta di ridisegno dei flussi per l’aggiornamento degli indirizzi Metodologia per la misurazione della qualità degli indirizzi Misurazione sperimentale della qualità degli indirizzi in tre archivi nazionali: Agenzia delle Entrate Camere di Commercio INPS 8 Data Quality and Data Privacy (Current) Joint Activity with University of Purdue, Indiana USA Publishing elementary data may violate privacy requirements, even when data are anonymized anonymization removes principal identifiers like SSN, Name+Surname+DOB, etc. Record matching privacy aware only the result of the intersection (AB) across data sets are shared and nothing else (not AAB and not B-AB) 9 Obiettivi ESTEEM Studio di problematiche di trust e qualità dei dati in sistemi P2P Specifica di sistemi di integrazione dati P2P con requisiti di trust Definizione di algoritmi di query processing quality- and trust-aware 10 P2P Systems P2P systems loosely coupled, dynamic, open Data sharing in such systems no centralized global schema peers mapping dynamically build new peers can make available new data schema 11 Data Quality EmployeeID Name Surname Salary Email arpa78 John Smith 2000 [email protected] eugi98 Edward Monroe 1500 [email protected] ghjk09 Anthony Wite 1250 [email protected] treg23 Marianne Collins 1150 [email protected] Attribute conflict EmployeeS1 Key conflict EmployeeID Name Surname Salary Email arpa78 John Smith 2600 [email protected] eugi98 Edward Monroe 1500 [email protected] ghjk09 Anthony White 1250 [email protected] dref43 Marianne Collins 1150 [email protected] EmployeeS2 12 Quality-aware query processing - 1 Key conflicts require the application of Record Matching techniques Attribute conflicts are solved by query time Conflict Resolution Techniques The resolution of such conflicts in P2P systems is an open issue: Definition of a quality-aware semantics for query answering in P2P systems Need to develop techniques for solving such conflicts according to the defined semantics 13 Quality-aware query processing - 2 Query language supporting the specification of conflict resolution strategies Important in P2P systems: research space pruning on the basis of quality characterization of sources 14 Privacy How to protect privacy when sharing data? With the source S1 and S2 issuing the Queries Q1 and Q2 respectively, at the end of the interaction S1 must learn result Q1 and nothing else S2 must learn result Q2 and nothing else Query Q1 S1 Result Q1 Query Q2 S2 Result Q2 15 Privacy-aware Record Matching - 1 A B AB Secure set intersection: (i) matching esatto; (ii) non di record; (iii) costosi Private data sharing: (i) matching esatto; (ii) schema un-aware 16 Privacy-aware Query Processing - 2 Algoritmi che consentano di fare privacy aware record matching in contesti P2P Problema della third party Prime proposte ElAbbadi ICDE 2006 ma matching esatto 17 Trust Trust typically associated to a source as a whole Need for finer level characterization Eg: Ministero delle Finanze affidabile rispetto ai Codici Fiscali 18 Modello di Trust per le sorgenti dati -1 Previous proposals: the whole organization (peer) Our proposal: <Organization, Data Type> # of <D, Orgk> complaints sent by Orgi R( Org k , C D ) n i i i, k , D i, k , D Org i O # of Dexchanges of Orgk 19 Modello di Trust per le sorgenti dati - 2 Drawback: Centralized Need for: Decentralized More flexible model (e.g. trust associated to views) 20 Modello di Trust per le sorgenti dati - 3 More general trust characterization based on the evaluation of a peer’s assertion on some metadata: Data quality-aware: trust computed on the basis of the declared quality of provided data Privacy-aware: trust computed on the basis of the declared privacy level different roles for providers and consumers: e.g. a provider can decide not to release data if a requester is not privacy - trusted (or to adopt specific technique) 21