A Method to Develop Description Logic Ontologies Iteratively Based on Competency Questions: an Implementation

Paper presented on Ontobras - 2013

Transcript

1. A Method to Develop Description Lo ic Ontolo ies Iteratively Based on Competency Questions: an Implementation Yuri Malheiros and Fred Freitas (CIn - UFPE) (DCE - UFPB - Campus IV)
2. 90’s 2013 From arcane art form to en ineerin
3. 90’s 2013 97 Methontolo y 01 OTKM 08 Ontolo y 101
4. 90’s 2013 01 WebODE 08 NeOn Toolkit 03 Proté é
5. The methodolo ies are sli htly di erent, but share common features. For example, iterative way of development and the use of competency questions
6. Problem for DL ontolo y en ineers
7. Includin axioms is a somewhat complex task, and most DL ontolo ies don’t include them! But, they are the core of DL ontolo ies
8. How to o further?
9. We are creatin a method and a implementation that reasons over Description Lo ic competency questions to build or evolve an ontolo y in an iterative way
10. ✓ Definition of requirements (CQs) ✓ Evaluation ✓ Automatic addition of new axioms (with user’s a reement)
11. Suppose this ontolo y: Herbivorous ≡ Animal ⊓ ∀eats.¬meat Cow ≡ Animal ⊓ ∀eats. rass
12. The user defines a competency question: Are cows herbivorous? And its expected answer: Yes.
13. Is it possible to answer the competency question correctly usin the previous axioms?
14. Is it possible to answer the competency question correctly usin the previous axioms? No.
15. It cannot infer: Cows ⊑ Herbivorous
16. The ontolo y needs more axioms to answer the competency question correctly. Then, the system should ask the user.
17. The system asks the user: Are rass and meat disjoint?
18. The user answer: Yes.
19. The system adds the followin axiom to the ontolo y: Grass ⊑ ¬Meat
20. Is it possible to answer the competency question correctly usin the previous axioms?
21. Is it possible to answer the competency question correctly usin the previous axioms? Yes.
22. The process 1. Load an ontolo y 2. Write a CQ and define the expected answer 3. Does the current DL ontolo y answers the question? If so, the system shows that the CQ is satisfied and it returns to step 2. 4. However, if the system cannot answer the CQ, it enerates a question for the user and asks for the answer to it. 5. The user answer the system’s question. 6. Return to the step 3
23. How does it work?
24. The system has three components: - Natural lan ua e query - Question enerator - Ontolo y builder
25. Natural lan ua e query It processes natural lan ua e competency questions and make queries
26. Natural lan ua e query It uses OWL API and HermiT OWL reasoner
27. is red wine a wine ? 1. separate the tokens
28. is red wine a wine ? 1. separate the tokens
29. is red wine a wine ? 2. apply a POS ta er vbz jj nn dt nn
30. 3. check the rules Is-a question: is (Noun|Adjective|Number)+ (a|an) (Noun|Adjective|Number)+ Property value question: (does|do) (Noun|Adjective|Number)+ have Noun (Noun|Adjective|Number)+ (does|do) (Noun|Adjective|Number)+ Verb (Noun|Adjective|Number)+ Existence question: which (Noun|Adjective|Number)+ exist which (Noun|Adjective|Number)+ have (Noun|Verb) (some|only)? (Noun| Adjective|Number)+
31. 3. check the rules Is-a question: is (Noun|Adjective|Number)+ (a|an) (Noun|Adjective|Number)+ is red wine a wine ? vbz jj nn dt nn
32. 4. identify entities Is-a question: is (Noun|Adjective|Number)+ (a|an) (Noun|Adjective|Number)+ is red wine a wine ? vbz jj nn dt nn
33. 4. identify entities Is-a question: is (Noun|Adjective|Number)+ (a|an) (Noun|Adjective|Number)+ is red wine a wine ? vbz jj nn dt nn
34. 4. identify entities Is-a question: is (Noun|Adjective|Number)+ (a|an) (Noun|Adjective|Number)+ is red wine a wine ? vbz jj nn dt nn
35. 4. identify entities Is-a question: is (Noun|Adjective|Number)+ (a|an) (Noun|Adjective|Number)+ is red wine a wine ? vbz jj nn dt nn RedWine
36. 4. identify entities Is-a question: is (Noun|Adjective|Number)+ (a|an) (Noun|Adjective|Number)+ is red wine a wine ? vbz jj nn dt nn RedWine Wine
37. The component runs the HermiT reasoner and uses OWL API to retrieve the answer 5. search for the answer
38. Question enerator This component is still bein studied, since it represents a new description lo ics problem
39. Question enerator Currently, we are assumin that an oracle for that problem exists. Then, the user provides the questions.
40. Ontolo y builder Add new axioms to an ontolo y combinin a system’s question and an answer automatically
41. Is-a system’s question: Example: Is red wine a wine? Answers: Yes, no, true or false Axiom: RedWine ⊑ Wine
42. Property value system’s question: Examples: Does bancroft chardonnay have only color white? Do birds eat rass? Answers: Yes, no, true or false Axioms: BancrofChardonnay ⊑ ∀hasColor.White Bird ⊑ ∃eat.Grass
43. Existence system’s question: Examples: Which wines exist? Answers: A list of classes separated by commas or “and”. For example, “red wine, white wine”. Axioms: RedWine ⊑ Wine WhiteWine ⊑ Wine
44. Results
45. We tested the natural lan ua e query component usin three ontolo ies: - wine.owl - pizza.owl - travel.owl
46. wine.owl is red wine a wine? true does bancroft chardonnay have color white? true
47. wine.owl which red wines exist? Beaujolais, CabernetFranc, CabernetSauvi non, Chianti, CotesDOr, DryRedWine, Mar aux, Medoc, Merita e, Merlot, Pauillac, PetiteSyrah, PinotNoir, Port, RedBordeaux, RedBur undy, RedTableWine, StEmilion, and Zinfandel which wines have su ar sweet? IceWine, LateHarvest, Port, Sauternes, and SweetRieslin
48. pizza.owl is napoletana a cheesey pizza? true which spicy pizzas exist? AmericanHot, Cajun, CheeseyVe etableToppin , IceCream, PolloAdAstra, and SloppyGiuseppe which meaty pizza has toppin some ham toppin ? Capricciosa, CheeseyVe etableToppin , IceCream, LaReine, Parmense, and Siciliana
49. travel.owl is yo a an activity? true does four seasons have ratin three star ratin ? true which adventures exist? BunjeeJumpin , and Safari which accommodations has ratin one star ratin ? Camp round, and Safari
50. To test the ontolo y builder component, we created a TestWine class in wine.owl, and answered some system’s questions
51. Question: is test wine a wine? Answer: true Code: Class: vin:TestWine SubClassOf: vin:Wine
52. Question: does test wine have only color white? Answer: true Code: Class: vin:TestWine SubClassOf: vin:Wine vin:hasColor only vin:White
53. Question: which test wines exist? Answer: red wine Code: Class: vin:RedWine EquivalentTo: vin:Wine and (vin:hasColor value vin:Red) SubClassOf: vin:TestWine
54. Conclusions
55. We developed a method and (partially) implemented a system to support a novel process for terminolo ical DL ontolo y buildin (TBox)
56. We have two components runnin : natural lan ua e query and ontolo y builder. Naturally, both need improvements: more complex CQs, more CQ types, etc.
57. Future work
58. We intend to: Build a complete tool to support this type of development, and the tool may be inte rated with popular ontolo y environments like Proté é
59. We intend to: Study the decidability and complexity of the new DL problem for the families of DL lan ua es Controlled tests with lar er ontolo ies for a fair evaluation
60. Thank You yuri@dce.ufpb.br