Presentation is loading. Please wait.

Presentation is loading. Please wait.

Mapping Ontology classes to Wordnet synsets

Similar presentations


Presentation on theme: "Mapping Ontology classes to Wordnet synsets"— Presentation transcript:

1 Mapping Ontology classes to Wordnet synsets
Xiangqian Li

2 Content Objectives Process Evaluation Construct taxonomies Matching
Mapping selection postprocessing Evaluation

3 Previously on SView Portal
搜索语义tags, 关键词搜索实体 go Selected Tags: location X city,metropolis X Categories: national capital national capital state capital provincial capital national capital others national capital state capital provincial capital

4 Objectives A semantic mapping from an ontology class C to a noun synset S indicates the relateness between two heterogeneous concepts in human senses. Map ontology classes to wordnet synsets Ontology class to Wordnet synset mappings Tag entities With semantic tags (synset) Find entities via Navigation with human senses

5 Objectives ... ... Uniform taxonomy derived from Wordnet
noun synsets with the hyponymy relations. We call it Synset Hyponymy Graph(SHG).

6 Process construct taxonomy for the ontology and corresponding SHG.
matching with multiple strategies Concept Cone Matcher Linguistic Matcher Similarity Flooding method based on SPG select mappings via homomorphism constraint Naive selection based on greedy strategy Select suitable strategies blocking skeleton selection supplement selection

7 Process / construct taxonomies
Ontology Classification: Use HermiT to inference undercovered subsumption relations underlying in the ontology. This process outputs a DAG, with node containing one class or several equivalent classes and edge representing subsumption relations. Load a synset hyponymy graph that the nodes are in the whole SHG and is homomorphic to the whole SHG.

8 Process / matching ConceptCone matcher
Construct virtual document of each concept in ontologies and Wordnet using Concept Cone. Caculate similarity using cosine value of two document vectors. current closer to the current node higher coefficient of term concept cone

9 Process / matching Linguistic Matcher
Compute the maximum similarity values between multiple labels of a class taxonomy node and the words in a synset taxonomy node as the similarity values of the two nodes using ISub. Combine the similarity of it to the similarity of the ConceptCone matcher using linear weighted sum.

10 Process / matching Similarity Flooding 1 in SPG:
Start a similarity propagation iteration till its convergence or predefined number of iterations reached. The intuition is: If two entities are similar, the entities having hypernymy or hyponymy relations with it is may be also similar. C1 map1 S1 map1 C2 S2 map2 map2 1 Melnik, Sergey, Hector Garcia-Molina, and Erhard Rahm. "Similarity flooding: A versatile graph matching algorithm and its application to schema matching." InProc. 18th ICDE Conf.(Best Student Paper award)

11 Process / Select mappings
Construct the SPG as the previous described steps. Divide it to several weakly connected components. For each component, evaluate its size. If it's too large, find the minimum cut edge set and remove these edges to partition this component to more smaller components.

12 Process / select mappings
First consider the component that covers the most nodes in the ontology taxonomy. selection via homomorphism constraint skeleton mappings component remove mappings that have the same source class node with the skeletons in all components or have conflicts with them

13 Process / select mappings
We require all ontology mappings have exactly one mapping to a wordnet synset. After building the skeleton mappings(anchors), we use several rules to select high quality mappings for the nodes haven’t been mapped.

14 Evaluation We conducted experiments on several popular ontologies, including schema.org, SUMO, DBpedia, UMBEL, GoodRelations, MusicOntology, Yago and several ontologies collected by Falcon in NJVR. The amount of the mappings are huge. We sampled a part of them to ask human to evaluate the quality of the mappings.

15 Evaluation Example

16 Evaluation Results

17 Thank you.


Download ppt "Mapping Ontology classes to Wordnet synsets"

Similar presentations


Ads by Google