Presentation is loading. Please wait.

Presentation is loading. Please wait.

Property consolidation for entity browsing

Similar presentations


Presentation on theme: "Property consolidation for entity browsing"— Presentation transcript:

1 Property consolidation for entity browsing
徐江

2 Scenario In fact, due to the fusion of entity descriptions from different sources, it frequently happens that many different properties from a variety of sources are semantically equivalent.

3 Property consolidation
Property consolidation, namely to consolidate a set of semantically equivalent properties into a feature. Currently, SView enables personalized property consolidation. How to automatically discover equivalent properties to alleviate user involvement ?

4 Related work Cheatham, Michelle, and Pascal Hitzler. "The Properties of Property Alignment." Proceedings OM-2014, The Ninth International Workshop on Ontology Matching, at the 13th International Semantic Web Conference, ISWC An in-depth exploration of the performance of current alignment systems on the only commonly accepted alignment benchmark that involves matches between properties. A benchmark involving properties is also proposed: YAGO-DBPedia core concept: either the first verb in the label that is greater than four characters long or, if there is no such verb, the first noun in the label, together with any adjectives that modify that noun.

5 Related work linguistic techniques used in ontology matching:
Edit-Distance-Based Strategy: isub, WordNet… Vector Distance (VD)-Based Strategy: - virtual document of a URIref declared in an ontology - Document similarity can be computed by traditional vector space techniques( tf*idf … )

6 Related work Existing approaches address entity coreference mainly from two directions: - equivalence inference mandated by OWL semantics (sameas, ifp, fp, cardinality) - similarity computation between property-value pairs

7 Related work Hu, Wei, Jianfeng Chen, and Yuzhong Qu. "A self-training approach for resolving object coreference on the semantic web." Proceedings of the 20th international conference on World wide web. ACM, 2011. For an RDF graph G, the matchability between two properties pi, pj in a kernel set D for an object URI is computed by:

8 Related work Gong, Saisai, Wei Hu, and Yuzhong Qu. "Leveraging Distributed Human Computation and Consensus Partition for Entity Coreference." The Semantic Web: Trends and Challenges. Springer International Publishing, Finds other entity descriptions that may be consolidated with it by using machine learning and will be confirmed by the user.

9 Matching schemas in online communities: A web 2.0 approach ICDE 2008
Why need interaction The fundamental reason is that matching is an inherently knowledge-intensive activity Challenges and Solution which questions to ask users Verify intermediate predictions bday is of type DATE Learn simple domain integrity constraints Verify final match predictions For complex matching indirectly evaluation evaluating community users’ reliability and combining their answers Simple trusted user and untrusted user How to post question mix user evaluation question with true question Post the same question to different users follow a distribution model

10 Matching schemas in online communities: A web 2.0 approach ICDE 2008

11 A hybrid machine-crowdsourcing system for matching web tables ICDE 2014
Challenge: What constitutes a “beneficial” column and should therefore be crowdsourced to determine the right concept for that column Solution Utility function Matching difficulty of a column Influence of column

12 A hybrid machine-crowdsourcing system for matching web tables ICDE 2014
α: prior probabilities of the machine 1−α: crowdsourcing influence

13 Large-scale Interactive Ontology Matching: Algorithms and Implementation ECAI 2012
it is crucial to reduce the number of questions to the human expert automatic decisions based on users’ feedback can significantly reduce the number of questions in practice Ambiguity Conflicts with semantic index

14 Overview Given an entity identified by a set of coreferent uris with properties and corresponding property values, get a list of features Three steps of the approach Discover matchable properties from their meta-information and values Highly confident features are to be confirmed by the user Collect all users’ results to optimize the model.

15 Prop consolidation wizard (1)

16 Prop consolidation wizard (2)

17 Thanks


Download ppt "Property consolidation for entity browsing"

Similar presentations


Ads by Google