LODQA Natural language query processing for SPARQL generation http://lodqa.dbcls.jp (demo system)

Jin-Dong Kim (DBCLS) Kevin Cohen (University of Colorado) Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

Motivation ● Linked

Open Data (LOD) increasing. ● SPARQL emerging as a query language. ● SPARQL is hard to write ✔

even for experts

● Natural

language query conversion to SPARQL can provide public users with a gentle interface layer.

Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

Demo ● LODQA ✔

prototype system

http://lodqa.dbcls.jp

Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

LODQA - two-step approach ● Pseudo ✔

SPARQL generation

Linguistic processing

● SPARQL ✔

instantiation

Adaptation to the target DB

Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

Pseudo SPARQL generation ● Analysis ✔ ✔ ✔

Parsing Base Noun Chunks (BNC) Targeting (typing) ➔ Find

the BNC modified by the wh-determiner, what.

● Generation ✔ ✔

(graph conversion)

Entity instantiation conditioning

Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

SPARQL instantiation ● URL ✔

lookup

OntoFinder (REST service) ➔ http://ontofinder.dbcls.jp ➔ Receives

a list of terms, and a (list of) ontology IDs ➔ Returns matching URIs to every term. ✔

“Kabuki Syndrom” in OMIM ➔ http://purl.bioontology.org/ontology/OMIM/147920

Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

SPARQL instantiation ● Typing ✔

?t1 :isa [genes] ➔ ?t1

umls:tui "T028"^^xsd:string

● Relation ✔

instantiation

?t1 [associated, with] ?t2 ➔ ?t1

?r1 ?t2. ➔ ?t1 ?r1 ?x1. ?x1 ?r2 ?t2.

?t1

?t2 Maximum sensitivity

?t1

?t2 ?x1

Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

SPARQL instantiation ● Strategy ✔

maximize the sensitivity first, ➔ Trivial,

✔

Generic

then, improve the specificity (filtering). ➔ Might

be more difficult ➔ Relation classification ➔ Mapping linguistic and schematic representation of relations

Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

LODQA ● Pseudo ✔ ✔ ✔ ✔

SPARQL generation

Base noun chunking Pattern Enju Parsing Targeting Pattern Conditioning Shortest path

● SPARQL ✔ ✔

instantiation

Entity instantiation Relation instantiation

OntoFinder Any path

granularity Filtering Classification

Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

Conclusions ● LODQA ✔

Is at its beginning stage ➔ We

are seeking a best starting point for open collaborative development.

✔

Toward an open source project. ➔ We

✔

are now writing unit-tests.

Prototype system is available ➔ http://lodqa.dbcls.jp

Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

Future directions 1 ● Parser ✔

Other parsers for other languages

● Graph ✔

Linguistic structure processing, e.g., coordinations, sortal coreferences.

● URL ✔ ✔ ✔

lookup

Matamap, Bioportal annotator for Bio-domain DBPedia spotlight for general domain Quantity expressions

● Target ✔ ✔

conversion

LODs

Bioportal LODs, Bio2RDF LODs DBpedia

Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

Future Direction 2 ● Federated

search

Language processing

answer Pseudo SPARQL aggregation

Adapdation to LODs

… Multiple LODs Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS

Future directions 3 ● To

make it an open source project ● To collect queries ✔ ✔

To direct linguistic changes to NLP community. To direct LOD challenges to LOD community.

Licensed under a Creative Commons Attribution 3.0 Unported License - DBCLS