Maintaining Information Integration Ontologies Alexandros Valarakos Georgios Paliouras

Maintaining Information Integration Ontologies Alexandros Valarakos, Georgios Paliouras, Vangelis Karkaletsis, Georgios Sigletos, Georgios Vouros Software & Knowledge Engineering Lab Inst. of Informatics & Telecommunications NCSR “Demokritos” http: //www. iit. demokritos. gr/skel DCAG, Ulm, December 6, 2003

Structure of the talk • • Information integration in CROSSMARC Semi-automated ontology enrichment Clustering “synonyms” Conclusions 6/12/200 Maintaining Information Integration DCAG, Ulm 2

CROSSMARC Objectives Develop technology for Information Integration that can: • crawl the Web for interesting Web pages, • extract information from pages of different sites without a standardized format (structured, semi-structured, free text), • process Web pages written in several languages, • be customized semi-automatically to new domains and languages, • deliver integrated information according to personalized profiles. 6/12/200 Maintaining Information Integration DCAG, Ulm 3

CROSSMARC Architecture Ontology 6/12/200 Maintaining Information Integration DCAG, Ulm 4

CROSSMARC Ontology • Meta-conceptual layer • Embodies domain-independent semantics • Conceptual layer • Contains relevant concepts of each domain • Instance layer • Contains relevant individuals of each domain • The lexical layer • Language dependent realizations of domain information 6/12/200 Maintaining Information Integration DCAG, Ulm 5

Intel Pentium III P 3 PIII " src="https://present5.com/presentation/e174e16b00c4db850a4cc4b26b3e98ed/image-6.jpg" alt="CROSSMARC Ontology Intel Pentium III P 3 PIII " /> CROSSMARC Ontology Intel Pentium III P 3 PIII … Laptops Lexicon Processor Processor Name Intel Pentium 3 … Όνομα Επεξεργαστή Ontology Greek Lexicon 6/12/200 Maintaining Information Integration DCAG, Ulm 6

Structure of the talk • • Information integration in CROSSMARC Semi-automated ontology enrichment Clustering “synonyms” Conclusions 6/12/200 Maintaining Information Integration DCAG, Ulm 7

Ontology Enrichment A-box T-box An ontology captures knowledge in a static way, as it is a Conceptualization snapshot of knowledge from a particular point of view that Instances governs a certain domain of interest in a specific time-period. Evolving nature of ontology Ontology Maintenance part of Ontology Enrichment 6/12/200 Maintaining Information Integration DCAG, Ulm 8

Ontology Enrichment • We concentrate on instances (knowledge of the domain of interest). • Highly evolving domain (e. g. laptop descriptions) – New Instances characterize new concepts. e. g. ‘Pentium 2’ is an instance that denotes a new concept if it doesn’t exist in the ontology. – New surface appearance of an instance. e. g. ‘PIII’ is a different surface appearance of ‘Intel Pentium 3’ • The poor performance of many Information Integration systems is due to their incapability to handle the evolving nature of the domain they cover. 6/12/200 Maintaining Information Integration DCAG, Ulm 9

Ontology Enrichment Annotating Corpus Using Domain Ontology machine learning Corpus Additional annotations Multi-Lingual Domain Ontology Enrichment / Population Information extraction Validation Domain Expert 6/12/200 Maintaining Information Integration DCAG, Ulm 10

Results: Annotation phase only 6/12/200 Maintaining Information Integration DCAG, Ulm 11

Results: Full enrichment cycle Initial Instances Target Instances Iter-0 Iter-1 Iter-2 Processor Name 3 15 3 4 3 Cdrom Speed 2 8 3 3 - Screen Resolution 2 7 0 - - RAM 2 8 5 0 - Processor Speed 4 12 7 0 - HDD 2 8 5 0 - Initial Instances Target Instances Iter-0 Iter-1 Iter-2 processor. Name 8 6 15 4 3 3 4 2 cdrom. Speed 5 6 8 3 2 - screen. Resolution Ram RAM Processor Speed HDD 3 5 4 6 6 9 4 6 7 7 8 8 12 12 8 8 2 2 3 2 6 2 3 2 1 0 0 0 - 1 6/12/200 Maintaining Information Integration 25% of the initial ontology 75% of the initial 50% of the initial ontology DCAG, Ulm 12

Structure of the talk • • Information integration in CROSSMARC Semi-automated ontology enrichment Clustering “synonyms” Conclusions 6/12/200 Maintaining Information Integration DCAG, Ulm 13

Enrichment with synonyms • The number of instances for validation increases with the size of the corpus and the ontology. • So far, only enrichment with instances that participate in the ‘instance of’ relationship has been supported. • There is a need for supporting the enrichment of the ‘synonymy’ relationship (in different languages and domains). We approach this problem using … ONTOLOGY LEARNING 6/12/200 Maintaining Information Integration DCAG, Ulm 14

Enrichment with synonyms • Discover automatically different surface appearances of an instance (CROSSMARC synonymy relationship). • Issues to be handled: Synonym : ‘Intel pentium 3’ - ‘Intel p. III’ Orthographical : ‘Intel p 3’ - ‘intell p 3’ Lexicographical : ‘Hewlett Packard’ - ‘HP’ Combination : ‘Intell Pentium 3’ - ‘P III’ 6/12/200 Maintaining Information Integration DCAG, Ulm 15

Compression-based Clustering • COCL (COmpression-based CLustering): a model based algorithm that discovers typographic similarities between strings (sequences of elements-letters) over an alphabet (ASCII characters) employing a new score function CCDiff. • CCDiff is defined as the difference in the code length of a cluster (i. e. , of its instances), when adding a candidate string. Huffman trees are used as models of the clusters. • COCL iteratively computes the CCDiff of each new string from each cluster implementing a hill-climbing search. The new string is added to the closest cluster, or a new cluster is created (threshold on CCDiff ). 6/12/200 Maintaining Information Integration DCAG, Ulm 16

Compression-based Clustering Given CLUSTERS and candidate INSTANCES while INSTANCES do for each instance in INSTANCES compute CCDiff for every cluster in CLUSTERS end for each select instance from INSTANCES that maximizes the difference between its two smallest CCDiff’s if min(CCDiff) of instance > threshold create new cluster assign instance to new cluster remove instance from INSTANCES calculate code model for the new cluster add new cluster to CLUSTERS else assign instance to cluster of min(CCDiff) remove instance from INSTANCES recalculate code model for the cluster 17 end while 6/12/200 Maintaining Information Integration DCAG, Ulm

Results - Evaluation • Concept Generation Scenario We hide incrementally one cluster at a time and measure the ability of the algorithm to discover the hidden clusters Recall : 100% Precision : 75% • Instance Matching Scenario Dataset characteristics Cluster’s Name Cluster’s Type Amd Processor Name 19 100 Intel Processor Name 8 19 100 Hewlett-Packard Manufacturer Name 3 50 23 95, 6 Fujitsu-Siemens Manufacturer Name 5 40 29 96, 5 30 34 94, 1 Windows 98 Operating System 10 Windows 2000 Operating System 3 Instances kept (%) Correct Accuracy (%) 90 3 100 80 11 100 70 15 60 Instances 6/12/200 Maintaining Information Integration DCAG, Ulm 18

Structure of the talk • • Information integration in CROSSMARC Semi-automated ontology enrichment Clustering “synonyms” Conclusions 6/12/200 Maintaining Information Integration DCAG, Ulm 19

Conclusions • CROSSMARC is a complete multi-lingual information integration system. • Ontology Maintenance is crucial in evolving domains. • Ontology Enrichment helps the adaptation of the system to new domains saving time and effort. • Machine-learning based information extraction can assist the discovery of new instances. • Compression-based clustering discovers string similarities that support the enrichment with different surface appearances of an instance (“synonyms”). 6/12/200 Maintaining Information Integration DCAG, Ulm 20

References 1) 2) 3) 4) 5) B. Hachey, C. Grover, V. Karkaletsis, A. Valarakos, M. T. Pazienza, M. Vindigni, E. Cartier, J. Coch, Use of Ontologies for Cross-lingual Information Management in the Web, In Proceedings of the Ontologies and Information Extraction International Workshop held as part of the EUROLAN 2003, Romania, July 28 - August 8, 2003 M. T. Pazienza, A. Stellato, M. Vindigni, A. Valarakos, V. Karkaletsis, Ontology Integration in a Multilingual e-Retail System, In Proceedings of the HCI International Conference, Volume 4, pp. 785 -789, Heraklion, Crete, Greece, June 22 -27 2003. A. Valarakos, G. Sigletos, V. Karkaletsis, G. Paliouras, A Methodology for Semantically Annotating a Corpus Using a Domain Ontology and Machine Learning, In RANLP, 2003 A. Valarakos, G. Sigletos, V. Karkaletsis, G. Paliouras, G. Vouros, A Methodology for Enriching a Multi-Lingual Domain Ontology using Machine Learning, In Proceedings of the 6 th ICGL workshop on Text Processing for Modern Greek: from Symbolic to Statistical Approaches, held as part of the 6 th International Conference in Greek Linguistics, Rethymno, Crete, 20 September, 2003. A. Valarakos, G. Paliouras, V. Karkaletsis, G. Vouros, A Name-Matching Algorithm for Ontology Enrichment, In Proceedings of the Hellenic Artificial Intelligence Conference (SETN’ 04), Samos, May, 2004. 6/12/200 Maintaining Information Integration DCAG, Ulm 21