Exploring the Future Potential of AI-Enabled Smartphone Processors
Computing k-rank Answers with Ontological CP-nets
1. COMPUTING
K-RANK ANSWERS
WITH ONTOLOGICAL CP-NETS
* University of Oxford
** Politecnico di Bari
Tommaso Di Noia**, Thomas Lukasiewicz *,
Maria Vanina Martinez*, Gerardo I Simari *,
Oana Tifrea-Marciuska *
PRUV 2014
12. Conditional Preferences (CP)-Nets
Ceteris paribus
All other things being equal
E.g. if I fly on business class and on a night arrival time then I
always prefer morning departure over night departure.
13. Reasoning with CP nets
¨ Worsening flip
¤ Changing value of a variable so that it is less
preferred in some outcome (direct dominance)
md ma e ≻ md ma b
md na e ≻ md ma e
14. Reasoning with CP nets
¨ Ordering on outcomes
¤ o1 is preferred to o2 (o1 ≻ o2) iff there is a sequence of
worsening flips from o1 to o2
¨ Partial order
¤ o1 and o2 can be incomparable
22. Query Answering
md ma e
md ma b md na e nd ma e
md na b nd na e
nd na b
nd ma b
Top-k Query Answering with Ontological CP-nets
Q(X,Y)→ ∃Z class(Z) ^ dTime(X) ^ aTime(Y)
Top-1 answer to Q: <md,ma>
o1
23. Query Answering
md ma e
md ma b md na e nd ma e
md na b nd na e
nd na b
nd ma b
Top-k Query Answering with Ontological CP-nets
Q(X,Y)→ ∃Z class(Z) ^ dTime(X) ^ aTime(Y)
Top-2 answers to Q: {<md,ma>,…}
o1
24. Query Answering
md ma b md na e nd ma e
md na b nd na e
nd na b
nd ma b
Top-k Query Answering with Ontological CP-nets
Q(X,Y)→ ∃Z class(Z) ^ dTime(X) ^ aTime(Y)
Top-2 answers to Q: {<md,ma>,…}
o2 o3
25. Query Answering
md na e nd ma e
md na b nd na e
nd na b
nd ma b
Top-k Query Answering with Ontological CP-nets
Q(X,Y)→ ∃Z class(Z) ^ dTime(X) ^ aTime(Y)
Top-2 answers to Q: {<md,ma>,<md,na>}
o3
o4
26. Query Answering
nd ma e
md na b nd na e
nd na b
nd ma b
Top-k Query Answering with Ontological CP-nets
Q(X,Y)→ ∃Z class(Z) ^ dTime(X) ^ aTime(Y)
Top-3 answers to Q: {<md,ma>,<md,na>,
<nd,ma>}
o4o5
27. Query Answering
Theorems
q Computing the k-rank is PSpace (resp. 2EXPTIME) complete for linear (resp.
guarded) Datalog+/- . Computing k-rank is data complete for PSPACE.
• TGD is guarded iff it contains an atom in its body that contains all universally quantified
variables.
σ1 :P(X)∧R(X,Y)∧Q(Y)→∃Z R(Y,Z) YES.
σ2 :R(X,Y)∧R(Y,Z)→R(X,Z) NO.
• TGD is linear: rules with one body-atom
σ3 :R(X,Y) →∃Z R(X,Z)
Top-k Query Answering with Ontological CP-nets
28. Query Answering
Theorems
If the CP-net is a polytree, the query has bounded width and the
Datalog+/- ontology is guarded or linear top-k can be done in polynomial
time.
Top-k Query Answering with Ontological CP-nets
32. Query Answering
md ma e
md ma b md na e nd ma e
md na b nd na e
nd na b
nd ma b
Top-2 answers to Q:
{<md,ma>,<md,na>}
Top-k Query Answering with Ontological CP-nets
Q(X,Y)→ ∃Z class(Z) ^ dTime(X) ^ aTime(Y)
Checked: {<md,ma,e>}
Outcomes: {<md,ma,e>,
<md,ma,b>, <md,na,e>}
Outcomes / Checked:
{<md,ma,b>, <md,na,e>}
33. Query Answering
md ma e
md ma b md na e nd ma e
md na b nd na e
nd na b
nd ma b
Top-3 answers to Q:
{<md,ma>,<md,na>…}
Q(X,Y)→ ∃Z class(Z) ^ dTime(X) ^ aTime(Y)
Top-k Query Answering with Ontological CP-nets
34. Query Answering
md ma e
md ma b md na e nd ma e
md na b nd na e
nd na b
nd ma b
Top-3 answers to Q:
{<md,ma>,<md,na>…}
Q(X,Y)→ ∃Z class(Z) ^ dTime(X) ^ aTime(Y)
Top-k Query Answering with Ontological CP-nets
35. Query Answering
md ma e
md ma b md na e nd ma e
nd na e
nd na b
nd ma b
Top-k Query Answering with Ontological CP-nets
md na b
Top-3 answers to Q:
{ <md,ma>,
<md,na>,
<nd,ma>}
Q(X,Y)→ ∃Z class(Z) ^ dTime(X) ^ aTime(Y)
36. Query Answering
md ma e
md ma b md na e nd ma e
nd na e
nd na b
nd ma b
Top-k Query Answering with Ontological CP-nets
md na b
Checked: {<md,ma,e>,
<md,ma,b>, <md,na,e>,
<md,na,b>, <nd,ma,b>}
Outcomes:
<md,ma,e>,
<md,ma,b>,
<md,na,e>,
<nd,ma,b>,
<md,na,b>}
Outcomes / Checked:
{<nd,ma,b>}
Top-3 answers to Q:
{ <md,ma>,
<md,na>,
<nd,ma>}
Q(X,Y)→ ∃Z class(Z) ^ dTime(X) ^ aTime(Y)
Editor's Notes
semantic data – where data can be extended or constrained using ontological information
Social data-- where data contains preferences, tags and connections between objects (i.e. Alice is a friend of
Bob).
Use the analogy of sql
Ranking answers for conjunctive queries to datalog+- ontologies based on preferences encoded as CP-nets
We chose Cp-nets because as we will see they provide natural and concise and flexible graphical representation of qualitative preferences
D+/- is mire expressive than Dl-lite and has more compact representation.
Tuple-generating dependencies
Compare with bayes network ( conditional tables and independence…)
CP-net is tight: on the one hand, CP-net outcomes are constrained by the ontology, and on the other hand, they directly inform how answers to CQs are ranked.
Outcomes are constraint by the ontology
Add defintion of skyline, k-rank asnwer and ordering in few slides
Add definition of top-k answer ( outcome restricted the answers) and use the iterative skyline definiton
Add definition of top-k answer ( outcome restricted the answers) and use the iterative skyline definiton
Add definition of top-k answer ( outcome restricted the answers) and use the iterative skyline definiton
Add definition of top-k answer ( outcome restricted the answers) and use the iterative skyline definiton
Add definition of top-k answer ( outcome restricted the answers) and use the iterative skyline definiton