SlideShare a Scribd company logo
Exception-enriched Rule
Learning from Knowledge
Graphs
Mohamed Gad-Elrab1, Daria Stepanova1, Jacopo Urbani 2, Gerhard Weikum1
1Max-Planck-Institut für Informatik, Saarland Informatics Campus, Germany
2 Vrije Universiteit Amsterdam, Amsterdam, The Netherlands
21st October 2016
Knowledge Graphs (KGs)
2
• Huge collection of < 𝑠𝑢𝑏𝑗𝑒𝑐𝑡, 𝑝𝑟𝑒𝑑𝑖𝑐𝑎𝑡𝑒, 𝑜𝑏𝑗𝑒𝑐𝑡 > triples
• Positive facts under Open World Assumption (OWA)
• Possibly incomplete and/or inaccurate
Mining Rules from KGs
3
Amsterdam
isMarriedToJohn Kate
Chicago
isMarriedToBrad Anna
Berlin
hasBrother
isMarriedToDave ClaraisMarriedToBob Alice
Researcher
Berlin Football
Mining Rules from KGs
4
𝑟: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛 𝑋, 𝑍 ← 𝑖𝑠𝑀𝑎𝑟𝑟𝑖𝑒𝑑𝑇𝑜 𝑋, 𝑌 , 𝑙𝑖𝑣𝑒𝑠𝐼𝑛(𝑌, 𝑍)
Amsterdam
isMarriedToJohn Kate
Chicago
isMarriedToBrad Anna
Berlin
hasBrother
isMarriedToDave ClaraisMarriedToBob Alice
Researcher
Berlin Football
[Galárraga et al., 2015]
Mining Rules from KGs
5
𝑟: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛 𝑋, 𝑍 ← 𝑖𝑠𝑀𝑎𝑟𝑟𝑖𝑒𝑑𝑇𝑜 𝑋, 𝑌 , 𝑙𝑖𝑣𝑒𝑠𝐼𝑛(𝑌, 𝑍)
Amsterdam
isMarriedToJohn Kate
Chicago
isMarriedToBrad Anna
Berlin
hasBrother
isMarriedToDave ClaraisMarriedToBob Alice
Researcher
Berlin Football
1 2
Our Goal
6
𝑟: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛 𝑋, 𝑍 ← 𝑖𝑠𝑀𝑎𝑟𝑟𝑖𝑒𝑑𝑇𝑜 𝑋, 𝑌 , 𝑙𝑖𝑣𝑒𝑠𝐼𝑛 𝑌, 𝑍 , 𝑛𝑜𝑡 𝑖𝑠𝐴(𝑋, 𝑟𝑒𝑠)
Amsterdam
isMarriedToJohn Kate
Chicago
isMarriedToBrad Anna
Berlin
hasBrother
isMarriedToDave ClaraisMarriedToBob Alice
Researcher
Berlin Football
1 2
Problem Statement
• Quality-based theory revision problem
• Given
• Knowledge graph 𝐾𝐺
• Set of Horn rules 𝑅 𝐻
• Find the nonmonotonic revision 𝑅 𝑁𝑀 of 𝑅 𝐻
• Maximize top-k avg. confidence
• Minimize conflicting prediction
7
Problem Statement
• Quality-based theory revision problem
• Given
• Knowledge graph 𝐾𝐺
• Set of Horn rules 𝑅 𝐻
• Find the nonmonotonic revision 𝑅 𝑁𝑀 of 𝑅 𝐻
• Maximize top-k avg. confidence
• Minimize conflicting prediction
8
Unknown
Problem Statement: Conflicting Predictions
• Defining conflicts
• Measuring conflicts (auxiliary rules)
9
𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋)
𝑟2
𝑎𝑢𝑥
: 𝑛𝑜𝑡_𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋)
𝑅 =
𝑟1: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← 𝑖𝑠𝐺𝑎𝑚𝑒 𝑋 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 (𝑋)
𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋)
{𝑖𝑠𝐺𝑎𝑚𝑒 𝑎 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 𝑎 , ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑎 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴 𝑎 }
Output: {(𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 , 𝑟2: 𝑛𝑜𝑡 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 )}
𝑎 =
Problem Statement: Conflicting Predictions
• Defining conflicts
• Measuring conflicts (auxiliary rules)
10
𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋)
𝑟2
𝑎𝑢𝑥
: 𝑛𝑜𝑡_𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋)
𝑅 =
𝑟1: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← 𝑖𝑠𝐺𝑎𝑚𝑒 𝑋 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 (𝑋)
𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋)
{𝑖𝑠𝐺𝑎𝑚𝑒 𝑎 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 𝑎 , ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑎 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴 𝑎 }
Output: {(𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 , 𝑟2: 𝑛𝑜𝑡 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 )}
𝑎 =
Problem Statement: Conflicting Predictions
• Defining conflicts
• Measuring conflicts (auxiliary rules)
11
𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋)
𝑟2
𝑎𝑢𝑥
: 𝑛𝑜𝑡_𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋)
𝑅 =
𝑟1: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← 𝑖𝑠𝐺𝑎𝑚𝑒 𝑋 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 (𝑋)
𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋)
{𝑖𝑠𝐺𝑎𝑚𝑒 𝑎 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 𝑎 , ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑎 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴 𝑎 }
(𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 , 𝑛𝑜𝑡_𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 )
𝑎 =
Approach Overview
Step 1
• Mining Horn Rules
Step 2
• Extracting Exception Witness Set (EWS)
Step 3
• Constructing Candidate Revisions
Step 4
• Selecting the Best Revision
12
bornInUSA livesInUSA stateless emigrant singer poet
p1 ✓ ✓
p2 ✓ ✓
p3 ✓ ✓
p4 ✓ ✓ ✓
p5 ✓ ✓ ✓
p6 ✓ ✓
p7 ✓ ✓
p8 ✓ ✓ ✓ ✓
p9 ✓ ✓ ✓
p10 ✓ ✓ ✓ ✓
p11 ✓ ✓ ✓
Step 1: Mining Horn Rules
13
𝒓𝒊: 𝒍𝒊𝒗𝒆𝒔𝑰𝒏𝑼𝑺𝑨 𝑿 ← 𝒃𝒐𝒓𝒏𝑰𝒏𝑼𝑺𝑨(𝑿)
NormalAb-normal
bornInUSA livesInUSA stateless emigrant singer poet
p1 ✓ ✓
p2 ✓ ✓
p3 ✓ ✓
p4 ✓ ✓ ✓
p5 ✓ ✓ ✓
p6 ✓ ✓
p7 ✓ ✓
p8 ✓ ✓ ✓ ✓
p9 ✓ ✓ ✓
p10 ✓ ✓ ✓ ✓
p11 ✓ ✓ ✓
𝒓𝒊: 𝒍𝒊𝒗𝒆𝒔𝑰𝒏𝑼𝑺𝑨 𝑿 ← 𝒃𝒐𝒓𝒏𝑰𝒏𝑼𝑺𝑨(𝑿)
NormalAb-normalStep 2: Extracting Exception Witness Set (EWS)
14
𝐸𝑊𝑆𝑖 = {𝑒𝑚𝑖𝑔𝑟𝑎𝑛𝑡 𝑋 , 𝑝𝑜𝑒𝑡 𝑋 }
Step 3: Constructing Candidate Revisions
• Horn rules
• Rule revisions
15
𝑟1: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛𝑈𝑆𝐴 𝑋 ← 𝑏𝑜𝑟𝑛𝐼𝑛𝑈𝑆𝐴 𝑋 , 𝑛𝑜𝑡 𝑝𝑜𝑒𝑡 𝑋
𝑟1: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛𝑈𝑆𝐴 𝑋 ← 𝑏𝑜𝑟𝑛𝐼𝑛𝑈𝑆𝐴 𝑋 , 𝑛𝑜𝑡 𝑒𝑚𝑖𝑔𝑟𝑎𝑛𝑡 𝑋
…
𝑅 =
𝑟1: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛𝑈𝑆𝐴 𝑋 ← 𝑏𝑜𝑟𝑛𝐼𝑛𝑈𝑆𝐴(𝑋)
𝑟2: 𝑒𝑚𝑖𝑔𝑟𝑎𝑛𝑡 𝑋 ← 𝑠𝑡𝑎𝑡𝑒𝑙𝑒𝑠𝑠(𝑋)
𝐸𝑊𝑆 = {𝑝𝑜𝑒𝑡 𝑋 , 𝑒𝑚𝑖𝑔𝑟𝑎𝑛𝑡 𝑋 , … }
𝐸𝑊𝑆 = {𝑒1 𝑋 , 𝑒2(𝑋), … }
Step 4: Selecting the Best Revision
Finding globally best revision is expensive!
• Naïve ranker
• For each rule, pick the revision that maximizes confidence
• Works in isolation from other rules
• Partial materialization ranker
• 𝐾𝐺s are incomplete!
• Augment the original 𝐾𝐺 with predictions of other rules
• Rank revisions on avg. confidence of the rule and its auxiliary.
16
• Partial materialization
Ranking Rule’s Revisions
17
bornInUSA livesInUSA stateless emigrant singer poet
p1 ✓ ✓
p2 ✓ ✓
p3 ✓ ✓
p4 ✓ ✓ ✓
p5 ✓ ✓ ✓
p6 ✓ ✓
p7 ✓ ✓
p8 ✓ ✓ ✓ ✓
p9 ✓ ✓ ✓
p10 ✓ ✓ ✓ ✓
p11 ✓ ✓ ✓
𝒓𝒊: 𝒍𝒊𝒗𝒆𝒔𝑰𝒏𝑼𝑺𝑨 𝑿 ← 𝒃𝒐𝒓𝒏𝑰𝒏𝑼𝑺𝑨(𝑿)
NormalAb-normal
• Partial materialization
bornInUSA livesInUSA stateless emigrant singer poet
p1 ✓ ✓
p2 ✓ ✓
p3 ✓ ✓
p4 ✓ ✓ ✓ ✓
p5 ✓ ✓ ✓ ✓
p6 ✓ ✓ ✓
p7 ✓ ✓ ✓
p8 ✓ ✓ ✓ ✓
p9 ✓ ✓ ✓ ✓ ✓
p10 ✓ ✓ ✓ ✓
p11 ✓ ✓ ✓ ✓
𝒓𝒊: 𝒍𝒊𝒗𝒆𝒔𝑰𝒏𝑼𝑺𝑨 𝑿 ← 𝒃𝒐𝒓𝒏𝑰𝒏𝑼𝑺𝑨(𝑿)
NormalAb-normalRanking Rule’s Revisions
18
Ranking Rule’s Revisions
• Ordered partial materialization ranker
• Only rules with higher quality
• Ordered weighted partial materialization ranker
• KG fact weight = 1
• Predicted facts inherit their weights from the rules
19
Experiments
• Ruleset quality
20
*Higher is better
0.60
0.65
0.70
0.75
0.80
20 40 60 80 100
Avg.Confidence
Top-K (%) Rules
YAGO3
Horn Naive Ordered & Weighted PM
0.8
0.82
0.84
0.86
0.88
0.9
0.92
0.94
20 40 60 80 100Avg.Confidence
Top-K (%) Rules
IMDB
Horn Naive Ordered & Weighted PM
Facts 10M 2M
Rules 10K 25K
Experiments
• Predictions consistency
21*Lower is better
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
200 400 600 800 1000
ConflictRatio
Top-K Rules
YAGO3
Naive Ordered & Weighted PM
0
0.1
0.2
0.3
0.4
0.5
200 400 600 800 1000ConflictRatio
Top-K Rules
IMDB
Naive Ordered & Weighted PM
Experiments
• Examples
22
𝑖𝑠𝑀𝑜𝑢𝑛𝑡𝑎𝑖𝑛(𝑋) ← 𝑖𝑠𝐼𝑛𝐴𝑢𝑠𝑡𝑟𝑖𝑎(𝑋), 𝑖𝑠𝐼𝑛𝐼𝑡𝑎𝑙𝑦 𝑋 , 𝑛𝑜𝑡 𝑖𝑠𝑅𝑖𝑣𝑒𝑟 (𝑋)
𝑖𝑠𝑃𝑜𝑙𝑖𝑡𝑂𝑓𝑈𝑆𝐴 𝑋 ← 𝑏𝑜𝑟𝑛𝐼𝑛𝑈𝑆𝐴 𝑋 , 𝑖𝑠𝐺𝑜𝑣 𝑋 , 𝑛𝑜𝑡 𝑖𝑠𝑃𝑜𝑙𝑖𝑡𝑃𝑢𝑒𝑟𝑡𝑜𝑅𝑖𝑐𝑜(𝑋)
𝑏𝑜𝑟𝑛𝐼𝑛𝑈𝑆𝐴 𝑋 ← 𝑎𝑐𝑡𝑒𝑑𝐼𝑛𝑀𝑜𝑣𝑖𝑒 𝑋 , 𝑐𝑟𝑒𝑎𝑡𝑒𝑑𝑀𝑜𝑣𝑖𝑒 𝑋 , 𝑛𝑜𝑡 𝑤𝑜𝑛𝐹𝑖𝑙𝑚𝑓𝑎𝑟𝑒(𝑋)
Summary
• Conclusion
• Quality-based theory revision under OWA
• Partial materialization for ranking revisions
• Comparison of ranking methods on real life KGs
• Outlook
• Extending to higher arity predicates
• Binary predicates [Tran et al., to appear ILP2016]
• Evidence from text corpora
• Exploiting partial completeness
23
References
• [Angiulli and Fassetti, 2014] Fabrizio Angiulli and Fabio Fassetti. Exploiting domain knowledge to
detect outliers. Data Min. Knowl. Discov., 28(2):519–568, 2014.
• [Dimopoulos and Kakas, 1995] Yannis Dimopoulos and Antonis C. Kakas. Learning non-monotonic
logic programs: Learning exceptions. In Machine Learning: ECML-95, 8th European Conference on
Machine Learning, Heraclion, Crete, Greece, April 25-27, 1995, Proceedings, pages 122–137, 1995.
• [Galarraga et al., 2015] Luis Galarraga, Christina Teflioudi, Katja Hose, and Fabian M. Suchanek.
Fast Rule Mining in Ontological Knowledge Bases with AMIE+. In VLDB Journal, 2015.
• [Law et al., 2015] Mark Law, Alessandra Russo, and Krysia Broda. The ILASP system for learning
answer set programs, 2015.
• [Leone et al., 2006] Nicola Leone, Gerald Pfeifer, Wolfgang Faber, Thomas Eiter, Georg Gottlob,
Simona Perri, and Francesco Scarcello. 2006. The DLV system for knowledge representation and
reasoning. ACM Trans. Comput. Logic 7, 3 (July 2006), 499-562.
• [Suzuki, 2006] Einoshin Suzuki. Data mining methods for discovering interesting exceptions from
an unsupervised table. J. UCS, 12(6):627–653, 2006.
• [Tran et al., 2016] Hai Dang Tran, Daria Stepanova, Mohamed H. Gad-Elrab, Francesca A. Lisi,
Gerhard Weikum. Towards Nonmonotonic Relational Learning from Knowledge Graphs. ILP2016,
London, UK, to appear.
• [Katzouris et al., 2015] Nikos Katzouris, Alexander Artikis, and Georgios Paliouras. Incremental
learning of event definitions with inductive logic programming. Machine Learning, 100(2-3):555–
585, 2015.
24
Related Work
• Learning nonmonotonic programs
• E.g., [Dimopoulos and Kakas, 1995], ILASP [Law et al., 2015],
ILED [Katzouris et al., 2015], etc.
• Outlier detection in logic programs
• E.g., [Angiulli and Fassetti, 2014], etc.
• Mining exception rules
• E.g., [Suzuki, 2006], etc.
25
Problem Statement: Ruleset Quality
• Independent Rule Measure (𝑟𝑚)
• Support: 𝑠𝑢𝑝𝑝 𝐻 ← 𝐵 = 𝑠𝑢𝑝𝑝(𝐻 ∪ 𝐵)
• Coverage: c𝑜𝑣 𝐻 ← 𝐵 = 𝑠𝑢𝑝𝑝(𝐵)
• Confidence: c𝑜𝑛𝑓 𝐻 ← 𝐵 =
𝑠𝑢𝑝𝑝(𝐻∪𝐵)
𝑠𝑢𝑝𝑝(𝐵)
• Lift: 𝑙𝑖𝑓𝑡 𝐻 ← 𝐵 =
𝑐𝑜𝑛𝑓(𝐻←𝐵)
𝑠𝑢𝑝𝑝(𝐻)
• …
• Average Ruleset Quality
26
𝑞 𝑟𝑚 𝑅 𝑁𝑀, 𝐺 =
𝑟 ∈𝑅 𝑁𝑀
𝑟𝑚(𝑟, 𝐺)
𝑅 𝑁𝑀
Problem Statement: Conflicting Predictions
• Measuring conflicts (auxiliary rules)
27
𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋)
𝑟2
𝑎𝑢𝑥
: 𝑛𝑜𝑡_𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋)
𝑞 𝑐𝑜𝑛𝑓𝑙𝑖𝑐𝑡 𝑅 𝑁𝑀, 𝐺 =
|{(𝑝 𝑎 , 𝑛𝑜𝑡_𝑝 𝑎 ), … }|
{𝑛𝑜𝑡_𝑝 𝑎 , … }
Propositionalization
• Unary predicates
28
Binary Predicates
• hasType(enstein, scientist)
• isMarriedTo(elsa, einstein)
• bornIn(einstein, um)
Unary
• isAScientist(einstein)
• isMarriedToEinstein(elsa)
• bornInUlm(einstein)
Abstraction
• isAScientist(einstein)
• isMarriedToScientist(elsa)
• bornInGermany(einstein)
Experiments
• Input
• Experiment statistics
29
YAGO3 IMDB
Input Facts 10M 2M
Horn Rules 10K 25K
Revised Rules 6K 22K
General-purpose KG Domain-specific KG (Movies)
Experiments
• Predictions assessment
• Run DLV on YAGO and 𝑅 𝐻 then 𝑅 𝑁𝑀 seperately
• Sample facts such that fact 𝑓 ∈ 𝑌𝐴𝐺𝑂 𝐻𝑌𝐴𝐺𝑂 𝑁𝑀
• 73% of the sampled facts were found to be erroneous
30
checked
predictions
Ranking Rule’s Revisions
• Partial materialization ranker
• Augment the original 𝐾𝐺 with predictions of other rules
• Rank revisions on Avg. confidence of the 𝑟 and 𝑟 𝑎𝑢𝑥
31
𝑠𝑐𝑜𝑟𝑒 𝑟𝑒, 𝐾𝐺∗ =
𝑐𝑜𝑛𝑓 𝑟𝑒, 𝐾𝐺∗ + 𝑐𝑜𝑛𝑓(𝑟𝑒
𝑎𝑢𝑥, 𝐾𝐺∗)
2
where 𝑟𝑒 is the rule 𝑟 with exception 𝑒 & 𝐾𝐺∗
is the augmented 𝐾𝐺.

More Related Content

Similar to Exception-enrcihed Rule Learning from Knowledge Graphs

Mini datathon
Mini datathonMini datathon
Mini datathon
Kunal Jain
 
DeepLearning
DeepLearningDeepLearning
DeepLearning
ShahzadAsgharArain
 
Ph d sem_1@iitm
Ph d sem_1@iitmPh d sem_1@iitm
Ph d sem_1@iitm
Vinu Ev
 
Argumentation in Artificial Intelligence: From Theory to Practice (Practice)
Argumentation in Artificial Intelligence: From Theory to Practice (Practice)Argumentation in Artificial Intelligence: From Theory to Practice (Practice)
Argumentation in Artificial Intelligence: From Theory to Practice (Practice)
Mauro Vallati
 
Investigations on Local Search based Hybrid Metaheuristics
Investigations on Local Search based Hybrid MetaheuristicsInvestigations on Local Search based Hybrid Metaheuristics
Investigations on Local Search based Hybrid Metaheuristics
Luca Di Gaspero
 
04-Data-Analysis-Overview.pptx
04-Data-Analysis-Overview.pptx04-Data-Analysis-Overview.pptx
04-Data-Analysis-Overview.pptx
Shree Shree
 
Big data
Big dataBig data
Big data
Zeeshan Khan
 
Big data
Big dataBig data
Big data
Big dataBig data
Big data
Harshit Namdev
 
Hadoop PDF
Hadoop PDFHadoop PDF
Hadoop PDF
1904saikrishna
 
Scheduling advertisements on a web page to maximize revenue
Scheduling advertisements on a web page to maximize revenueScheduling advertisements on a web page to maximize revenue
Scheduling advertisements on a web page to maximize revenue
Shu-Jeng Hsieh
 
Performance OR Capacity #CMGimPACt2016
Performance OR Capacity #CMGimPACt2016 Performance OR Capacity #CMGimPACt2016
Performance OR Capacity #CMGimPACt2016
Alex Gilgur
 
Unveiling the properties of structured grammatical evolution
Unveiling the properties of structured grammatical evolutionUnveiling the properties of structured grammatical evolution
Unveiling the properties of structured grammatical evolution
Facultad de Informática UCM
 
Monte Carlo Tree Search for the Super Mario Bros
Monte Carlo Tree Search for the Super Mario BrosMonte Carlo Tree Search for the Super Mario Bros
Monte Carlo Tree Search for the Super Mario Bros
Chih-Sheng Lin
 
Skillwise Big data
Skillwise Big dataSkillwise Big data
Skillwise Big data
Skillwise Group
 
The Power of Probabilistic Thinking (keynote talk at ASE 2016)
The Power of Probabilistic Thinking (keynote talk at ASE 2016)The Power of Probabilistic Thinking (keynote talk at ASE 2016)
The Power of Probabilistic Thinking (keynote talk at ASE 2016)
David Rosenblum
 
NPS_TDA_forPDF_JPrendki
NPS_TDA_forPDF_JPrendkiNPS_TDA_forPDF_JPrendki
NPS_TDA_forPDF_JPrendki
Jennifer Prendki
 
Self-supervised Learning Lecture Note
Self-supervised Learning Lecture NoteSelf-supervised Learning Lecture Note
Self-supervised Learning Lecture Note
Sangwoo Mo
 
Pontificating quantification
Pontificating quantificationPontificating quantification
Pontificating quantification
Aaron Bedra
 
Stacked Ensembles in H2O
Stacked Ensembles in H2OStacked Ensembles in H2O
Stacked Ensembles in H2O
Sri Ambati
 

Similar to Exception-enrcihed Rule Learning from Knowledge Graphs (20)

Mini datathon
Mini datathonMini datathon
Mini datathon
 
DeepLearning
DeepLearningDeepLearning
DeepLearning
 
Ph d sem_1@iitm
Ph d sem_1@iitmPh d sem_1@iitm
Ph d sem_1@iitm
 
Argumentation in Artificial Intelligence: From Theory to Practice (Practice)
Argumentation in Artificial Intelligence: From Theory to Practice (Practice)Argumentation in Artificial Intelligence: From Theory to Practice (Practice)
Argumentation in Artificial Intelligence: From Theory to Practice (Practice)
 
Investigations on Local Search based Hybrid Metaheuristics
Investigations on Local Search based Hybrid MetaheuristicsInvestigations on Local Search based Hybrid Metaheuristics
Investigations on Local Search based Hybrid Metaheuristics
 
04-Data-Analysis-Overview.pptx
04-Data-Analysis-Overview.pptx04-Data-Analysis-Overview.pptx
04-Data-Analysis-Overview.pptx
 
Big data
Big dataBig data
Big data
 
Big data
Big dataBig data
Big data
 
Big data
Big dataBig data
Big data
 
Hadoop PDF
Hadoop PDFHadoop PDF
Hadoop PDF
 
Scheduling advertisements on a web page to maximize revenue
Scheduling advertisements on a web page to maximize revenueScheduling advertisements on a web page to maximize revenue
Scheduling advertisements on a web page to maximize revenue
 
Performance OR Capacity #CMGimPACt2016
Performance OR Capacity #CMGimPACt2016 Performance OR Capacity #CMGimPACt2016
Performance OR Capacity #CMGimPACt2016
 
Unveiling the properties of structured grammatical evolution
Unveiling the properties of structured grammatical evolutionUnveiling the properties of structured grammatical evolution
Unveiling the properties of structured grammatical evolution
 
Monte Carlo Tree Search for the Super Mario Bros
Monte Carlo Tree Search for the Super Mario BrosMonte Carlo Tree Search for the Super Mario Bros
Monte Carlo Tree Search for the Super Mario Bros
 
Skillwise Big data
Skillwise Big dataSkillwise Big data
Skillwise Big data
 
The Power of Probabilistic Thinking (keynote talk at ASE 2016)
The Power of Probabilistic Thinking (keynote talk at ASE 2016)The Power of Probabilistic Thinking (keynote talk at ASE 2016)
The Power of Probabilistic Thinking (keynote talk at ASE 2016)
 
NPS_TDA_forPDF_JPrendki
NPS_TDA_forPDF_JPrendkiNPS_TDA_forPDF_JPrendki
NPS_TDA_forPDF_JPrendki
 
Self-supervised Learning Lecture Note
Self-supervised Learning Lecture NoteSelf-supervised Learning Lecture Note
Self-supervised Learning Lecture Note
 
Pontificating quantification
Pontificating quantificationPontificating quantification
Pontificating quantification
 
Stacked Ensembles in H2O
Stacked Ensembles in H2OStacked Ensembles in H2O
Stacked Ensembles in H2O
 

Recently uploaded

Essentials of Automations: The Art of Triggers and Actions in FME
Essentials of Automations: The Art of Triggers and Actions in FMEEssentials of Automations: The Art of Triggers and Actions in FME
Essentials of Automations: The Art of Triggers and Actions in FME
Safe Software
 
GraphSummit Paris - The art of the possible with Graph Technology
GraphSummit Paris - The art of the possible with Graph TechnologyGraphSummit Paris - The art of the possible with Graph Technology
GraphSummit Paris - The art of the possible with Graph Technology
Neo4j
 
AI Fusion Buddy Review: Brand New, Groundbreaking Gemini-Powered AI App
AI Fusion Buddy Review: Brand New, Groundbreaking Gemini-Powered AI AppAI Fusion Buddy Review: Brand New, Groundbreaking Gemini-Powered AI App
AI Fusion Buddy Review: Brand New, Groundbreaking Gemini-Powered AI App
Google
 
Hand Rolled Applicative User Validation Code Kata
Hand Rolled Applicative User ValidationCode KataHand Rolled Applicative User ValidationCode Kata
Hand Rolled Applicative User Validation Code Kata
Philip Schwarz
 
OpenMetadata Community Meeting - 5th June 2024
OpenMetadata Community Meeting - 5th June 2024OpenMetadata Community Meeting - 5th June 2024
OpenMetadata Community Meeting - 5th June 2024
OpenMetadata
 
Vitthal Shirke Java Microservices Resume.pdf
Vitthal Shirke Java Microservices Resume.pdfVitthal Shirke Java Microservices Resume.pdf
Vitthal Shirke Java Microservices Resume.pdf
Vitthal Shirke
 
Microservice Teams - How the cloud changes the way we work
Microservice Teams - How the cloud changes the way we workMicroservice Teams - How the cloud changes the way we work
Microservice Teams - How the cloud changes the way we work
Sven Peters
 
LORRAINE ANDREI_LEQUIGAN_HOW TO USE ZOOM
LORRAINE ANDREI_LEQUIGAN_HOW TO USE ZOOMLORRAINE ANDREI_LEQUIGAN_HOW TO USE ZOOM
LORRAINE ANDREI_LEQUIGAN_HOW TO USE ZOOM
lorraineandreiamcidl
 
Need for Speed: Removing speed bumps from your Symfony projects ⚡️
Need for Speed: Removing speed bumps from your Symfony projects ⚡️Need for Speed: Removing speed bumps from your Symfony projects ⚡️
Need for Speed: Removing speed bumps from your Symfony projects ⚡️
Łukasz Chruściel
 
LORRAINE ANDREI_LEQUIGAN_HOW TO USE WHATSAPP.pptx
LORRAINE ANDREI_LEQUIGAN_HOW TO USE WHATSAPP.pptxLORRAINE ANDREI_LEQUIGAN_HOW TO USE WHATSAPP.pptx
LORRAINE ANDREI_LEQUIGAN_HOW TO USE WHATSAPP.pptx
lorraineandreiamcidl
 
Mobile App Development Company In Noida | Drona Infotech
Mobile App Development Company In Noida | Drona InfotechMobile App Development Company In Noida | Drona Infotech
Mobile App Development Company In Noida | Drona Infotech
Drona Infotech
 
KuberTENes Birthday Bash Guadalajara - Introducción a Argo CD
KuberTENes Birthday Bash Guadalajara - Introducción a Argo CDKuberTENes Birthday Bash Guadalajara - Introducción a Argo CD
KuberTENes Birthday Bash Guadalajara - Introducción a Argo CD
rodomar2
 
How to write a program in any programming language
How to write a program in any programming languageHow to write a program in any programming language
How to write a program in any programming language
Rakesh Kumar R
 
What is Augmented Reality Image Tracking
What is Augmented Reality Image TrackingWhat is Augmented Reality Image Tracking
What is Augmented Reality Image Tracking
pavan998932
 
Artificia Intellicence and XPath Extension Functions
Artificia Intellicence and XPath Extension FunctionsArtificia Intellicence and XPath Extension Functions
Artificia Intellicence and XPath Extension Functions
Octavian Nadolu
 
ALGIT - Assembly Line for Green IT - Numbers, Data, Facts
ALGIT - Assembly Line for Green IT - Numbers, Data, FactsALGIT - Assembly Line for Green IT - Numbers, Data, Facts
ALGIT - Assembly Line for Green IT - Numbers, Data, Facts
Green Software Development
 
DDS-Security 1.2 - What's New? Stronger security for long-running systems
DDS-Security 1.2 - What's New? Stronger security for long-running systemsDDS-Security 1.2 - What's New? Stronger security for long-running systems
DDS-Security 1.2 - What's New? Stronger security for long-running systems
Gerardo Pardo-Castellote
 
socradar-q1-2024-aviation-industry-report.pdf
socradar-q1-2024-aviation-industry-report.pdfsocradar-q1-2024-aviation-industry-report.pdf
socradar-q1-2024-aviation-industry-report.pdf
SOCRadar
 
Empowering Growth with Best Software Development Company in Noida - Deuglo
Empowering Growth with Best Software  Development Company in Noida - DeugloEmpowering Growth with Best Software  Development Company in Noida - Deuglo
Empowering Growth with Best Software Development Company in Noida - Deuglo
Deuglo Infosystem Pvt Ltd
 
Revolutionizing Visual Effects Mastering AI Face Swaps.pdf
Revolutionizing Visual Effects Mastering AI Face Swaps.pdfRevolutionizing Visual Effects Mastering AI Face Swaps.pdf
Revolutionizing Visual Effects Mastering AI Face Swaps.pdf
Undress Baby
 

Recently uploaded (20)

Essentials of Automations: The Art of Triggers and Actions in FME
Essentials of Automations: The Art of Triggers and Actions in FMEEssentials of Automations: The Art of Triggers and Actions in FME
Essentials of Automations: The Art of Triggers and Actions in FME
 
GraphSummit Paris - The art of the possible with Graph Technology
GraphSummit Paris - The art of the possible with Graph TechnologyGraphSummit Paris - The art of the possible with Graph Technology
GraphSummit Paris - The art of the possible with Graph Technology
 
AI Fusion Buddy Review: Brand New, Groundbreaking Gemini-Powered AI App
AI Fusion Buddy Review: Brand New, Groundbreaking Gemini-Powered AI AppAI Fusion Buddy Review: Brand New, Groundbreaking Gemini-Powered AI App
AI Fusion Buddy Review: Brand New, Groundbreaking Gemini-Powered AI App
 
Hand Rolled Applicative User Validation Code Kata
Hand Rolled Applicative User ValidationCode KataHand Rolled Applicative User ValidationCode Kata
Hand Rolled Applicative User Validation Code Kata
 
OpenMetadata Community Meeting - 5th June 2024
OpenMetadata Community Meeting - 5th June 2024OpenMetadata Community Meeting - 5th June 2024
OpenMetadata Community Meeting - 5th June 2024
 
Vitthal Shirke Java Microservices Resume.pdf
Vitthal Shirke Java Microservices Resume.pdfVitthal Shirke Java Microservices Resume.pdf
Vitthal Shirke Java Microservices Resume.pdf
 
Microservice Teams - How the cloud changes the way we work
Microservice Teams - How the cloud changes the way we workMicroservice Teams - How the cloud changes the way we work
Microservice Teams - How the cloud changes the way we work
 
LORRAINE ANDREI_LEQUIGAN_HOW TO USE ZOOM
LORRAINE ANDREI_LEQUIGAN_HOW TO USE ZOOMLORRAINE ANDREI_LEQUIGAN_HOW TO USE ZOOM
LORRAINE ANDREI_LEQUIGAN_HOW TO USE ZOOM
 
Need for Speed: Removing speed bumps from your Symfony projects ⚡️
Need for Speed: Removing speed bumps from your Symfony projects ⚡️Need for Speed: Removing speed bumps from your Symfony projects ⚡️
Need for Speed: Removing speed bumps from your Symfony projects ⚡️
 
LORRAINE ANDREI_LEQUIGAN_HOW TO USE WHATSAPP.pptx
LORRAINE ANDREI_LEQUIGAN_HOW TO USE WHATSAPP.pptxLORRAINE ANDREI_LEQUIGAN_HOW TO USE WHATSAPP.pptx
LORRAINE ANDREI_LEQUIGAN_HOW TO USE WHATSAPP.pptx
 
Mobile App Development Company In Noida | Drona Infotech
Mobile App Development Company In Noida | Drona InfotechMobile App Development Company In Noida | Drona Infotech
Mobile App Development Company In Noida | Drona Infotech
 
KuberTENes Birthday Bash Guadalajara - Introducción a Argo CD
KuberTENes Birthday Bash Guadalajara - Introducción a Argo CDKuberTENes Birthday Bash Guadalajara - Introducción a Argo CD
KuberTENes Birthday Bash Guadalajara - Introducción a Argo CD
 
How to write a program in any programming language
How to write a program in any programming languageHow to write a program in any programming language
How to write a program in any programming language
 
What is Augmented Reality Image Tracking
What is Augmented Reality Image TrackingWhat is Augmented Reality Image Tracking
What is Augmented Reality Image Tracking
 
Artificia Intellicence and XPath Extension Functions
Artificia Intellicence and XPath Extension FunctionsArtificia Intellicence and XPath Extension Functions
Artificia Intellicence and XPath Extension Functions
 
ALGIT - Assembly Line for Green IT - Numbers, Data, Facts
ALGIT - Assembly Line for Green IT - Numbers, Data, FactsALGIT - Assembly Line for Green IT - Numbers, Data, Facts
ALGIT - Assembly Line for Green IT - Numbers, Data, Facts
 
DDS-Security 1.2 - What's New? Stronger security for long-running systems
DDS-Security 1.2 - What's New? Stronger security for long-running systemsDDS-Security 1.2 - What's New? Stronger security for long-running systems
DDS-Security 1.2 - What's New? Stronger security for long-running systems
 
socradar-q1-2024-aviation-industry-report.pdf
socradar-q1-2024-aviation-industry-report.pdfsocradar-q1-2024-aviation-industry-report.pdf
socradar-q1-2024-aviation-industry-report.pdf
 
Empowering Growth with Best Software Development Company in Noida - Deuglo
Empowering Growth with Best Software  Development Company in Noida - DeugloEmpowering Growth with Best Software  Development Company in Noida - Deuglo
Empowering Growth with Best Software Development Company in Noida - Deuglo
 
Revolutionizing Visual Effects Mastering AI Face Swaps.pdf
Revolutionizing Visual Effects Mastering AI Face Swaps.pdfRevolutionizing Visual Effects Mastering AI Face Swaps.pdf
Revolutionizing Visual Effects Mastering AI Face Swaps.pdf
 

Exception-enrcihed Rule Learning from Knowledge Graphs

  • 1. Exception-enriched Rule Learning from Knowledge Graphs Mohamed Gad-Elrab1, Daria Stepanova1, Jacopo Urbani 2, Gerhard Weikum1 1Max-Planck-Institut für Informatik, Saarland Informatics Campus, Germany 2 Vrije Universiteit Amsterdam, Amsterdam, The Netherlands 21st October 2016
  • 2. Knowledge Graphs (KGs) 2 • Huge collection of < 𝑠𝑢𝑏𝑗𝑒𝑐𝑡, 𝑝𝑟𝑒𝑑𝑖𝑐𝑎𝑡𝑒, 𝑜𝑏𝑗𝑒𝑐𝑡 > triples • Positive facts under Open World Assumption (OWA) • Possibly incomplete and/or inaccurate
  • 3. Mining Rules from KGs 3 Amsterdam isMarriedToJohn Kate Chicago isMarriedToBrad Anna Berlin hasBrother isMarriedToDave ClaraisMarriedToBob Alice Researcher Berlin Football
  • 4. Mining Rules from KGs 4 𝑟: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛 𝑋, 𝑍 ← 𝑖𝑠𝑀𝑎𝑟𝑟𝑖𝑒𝑑𝑇𝑜 𝑋, 𝑌 , 𝑙𝑖𝑣𝑒𝑠𝐼𝑛(𝑌, 𝑍) Amsterdam isMarriedToJohn Kate Chicago isMarriedToBrad Anna Berlin hasBrother isMarriedToDave ClaraisMarriedToBob Alice Researcher Berlin Football [Galárraga et al., 2015]
  • 5. Mining Rules from KGs 5 𝑟: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛 𝑋, 𝑍 ← 𝑖𝑠𝑀𝑎𝑟𝑟𝑖𝑒𝑑𝑇𝑜 𝑋, 𝑌 , 𝑙𝑖𝑣𝑒𝑠𝐼𝑛(𝑌, 𝑍) Amsterdam isMarriedToJohn Kate Chicago isMarriedToBrad Anna Berlin hasBrother isMarriedToDave ClaraisMarriedToBob Alice Researcher Berlin Football 1 2
  • 6. Our Goal 6 𝑟: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛 𝑋, 𝑍 ← 𝑖𝑠𝑀𝑎𝑟𝑟𝑖𝑒𝑑𝑇𝑜 𝑋, 𝑌 , 𝑙𝑖𝑣𝑒𝑠𝐼𝑛 𝑌, 𝑍 , 𝑛𝑜𝑡 𝑖𝑠𝐴(𝑋, 𝑟𝑒𝑠) Amsterdam isMarriedToJohn Kate Chicago isMarriedToBrad Anna Berlin hasBrother isMarriedToDave ClaraisMarriedToBob Alice Researcher Berlin Football 1 2
  • 7. Problem Statement • Quality-based theory revision problem • Given • Knowledge graph 𝐾𝐺 • Set of Horn rules 𝑅 𝐻 • Find the nonmonotonic revision 𝑅 𝑁𝑀 of 𝑅 𝐻 • Maximize top-k avg. confidence • Minimize conflicting prediction 7
  • 8. Problem Statement • Quality-based theory revision problem • Given • Knowledge graph 𝐾𝐺 • Set of Horn rules 𝑅 𝐻 • Find the nonmonotonic revision 𝑅 𝑁𝑀 of 𝑅 𝐻 • Maximize top-k avg. confidence • Minimize conflicting prediction 8 Unknown
  • 9. Problem Statement: Conflicting Predictions • Defining conflicts • Measuring conflicts (auxiliary rules) 9 𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋) 𝑟2 𝑎𝑢𝑥 : 𝑛𝑜𝑡_𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋) 𝑅 = 𝑟1: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← 𝑖𝑠𝐺𝑎𝑚𝑒 𝑋 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 (𝑋) 𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋) {𝑖𝑠𝐺𝑎𝑚𝑒 𝑎 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 𝑎 , ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑎 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴 𝑎 } Output: {(𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 , 𝑟2: 𝑛𝑜𝑡 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 )} 𝑎 =
  • 10. Problem Statement: Conflicting Predictions • Defining conflicts • Measuring conflicts (auxiliary rules) 10 𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋) 𝑟2 𝑎𝑢𝑥 : 𝑛𝑜𝑡_𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋) 𝑅 = 𝑟1: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← 𝑖𝑠𝐺𝑎𝑚𝑒 𝑋 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 (𝑋) 𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋) {𝑖𝑠𝐺𝑎𝑚𝑒 𝑎 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 𝑎 , ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑎 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴 𝑎 } Output: {(𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 , 𝑟2: 𝑛𝑜𝑡 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 )} 𝑎 =
  • 11. Problem Statement: Conflicting Predictions • Defining conflicts • Measuring conflicts (auxiliary rules) 11 𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋) 𝑟2 𝑎𝑢𝑥 : 𝑛𝑜𝑡_𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋) 𝑅 = 𝑟1: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← 𝑖𝑠𝐺𝑎𝑚𝑒 𝑋 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 (𝑋) 𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋) {𝑖𝑠𝐺𝑎𝑚𝑒 𝑎 , 𝑖𝑠𝐵𝑎𝑠𝑒𝑑𝑂𝑛𝐽𝑃𝐴𝑛𝑖𝑚𝑒 𝑎 , ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑎 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴 𝑎 } (𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 , 𝑛𝑜𝑡_𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑎 ) 𝑎 =
  • 12. Approach Overview Step 1 • Mining Horn Rules Step 2 • Extracting Exception Witness Set (EWS) Step 3 • Constructing Candidate Revisions Step 4 • Selecting the Best Revision 12
  • 13. bornInUSA livesInUSA stateless emigrant singer poet p1 ✓ ✓ p2 ✓ ✓ p3 ✓ ✓ p4 ✓ ✓ ✓ p5 ✓ ✓ ✓ p6 ✓ ✓ p7 ✓ ✓ p8 ✓ ✓ ✓ ✓ p9 ✓ ✓ ✓ p10 ✓ ✓ ✓ ✓ p11 ✓ ✓ ✓ Step 1: Mining Horn Rules 13 𝒓𝒊: 𝒍𝒊𝒗𝒆𝒔𝑰𝒏𝑼𝑺𝑨 𝑿 ← 𝒃𝒐𝒓𝒏𝑰𝒏𝑼𝑺𝑨(𝑿) NormalAb-normal
  • 14. bornInUSA livesInUSA stateless emigrant singer poet p1 ✓ ✓ p2 ✓ ✓ p3 ✓ ✓ p4 ✓ ✓ ✓ p5 ✓ ✓ ✓ p6 ✓ ✓ p7 ✓ ✓ p8 ✓ ✓ ✓ ✓ p9 ✓ ✓ ✓ p10 ✓ ✓ ✓ ✓ p11 ✓ ✓ ✓ 𝒓𝒊: 𝒍𝒊𝒗𝒆𝒔𝑰𝒏𝑼𝑺𝑨 𝑿 ← 𝒃𝒐𝒓𝒏𝑰𝒏𝑼𝑺𝑨(𝑿) NormalAb-normalStep 2: Extracting Exception Witness Set (EWS) 14 𝐸𝑊𝑆𝑖 = {𝑒𝑚𝑖𝑔𝑟𝑎𝑛𝑡 𝑋 , 𝑝𝑜𝑒𝑡 𝑋 }
  • 15. Step 3: Constructing Candidate Revisions • Horn rules • Rule revisions 15 𝑟1: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛𝑈𝑆𝐴 𝑋 ← 𝑏𝑜𝑟𝑛𝐼𝑛𝑈𝑆𝐴 𝑋 , 𝑛𝑜𝑡 𝑝𝑜𝑒𝑡 𝑋 𝑟1: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛𝑈𝑆𝐴 𝑋 ← 𝑏𝑜𝑟𝑛𝐼𝑛𝑈𝑆𝐴 𝑋 , 𝑛𝑜𝑡 𝑒𝑚𝑖𝑔𝑟𝑎𝑛𝑡 𝑋 … 𝑅 = 𝑟1: 𝑙𝑖𝑣𝑒𝑠𝐼𝑛𝑈𝑆𝐴 𝑋 ← 𝑏𝑜𝑟𝑛𝐼𝑛𝑈𝑆𝐴(𝑋) 𝑟2: 𝑒𝑚𝑖𝑔𝑟𝑎𝑛𝑡 𝑋 ← 𝑠𝑡𝑎𝑡𝑒𝑙𝑒𝑠𝑠(𝑋) 𝐸𝑊𝑆 = {𝑝𝑜𝑒𝑡 𝑋 , 𝑒𝑚𝑖𝑔𝑟𝑎𝑛𝑡 𝑋 , … } 𝐸𝑊𝑆 = {𝑒1 𝑋 , 𝑒2(𝑋), … }
  • 16. Step 4: Selecting the Best Revision Finding globally best revision is expensive! • Naïve ranker • For each rule, pick the revision that maximizes confidence • Works in isolation from other rules • Partial materialization ranker • 𝐾𝐺s are incomplete! • Augment the original 𝐾𝐺 with predictions of other rules • Rank revisions on avg. confidence of the rule and its auxiliary. 16
  • 17. • Partial materialization Ranking Rule’s Revisions 17 bornInUSA livesInUSA stateless emigrant singer poet p1 ✓ ✓ p2 ✓ ✓ p3 ✓ ✓ p4 ✓ ✓ ✓ p5 ✓ ✓ ✓ p6 ✓ ✓ p7 ✓ ✓ p8 ✓ ✓ ✓ ✓ p9 ✓ ✓ ✓ p10 ✓ ✓ ✓ ✓ p11 ✓ ✓ ✓ 𝒓𝒊: 𝒍𝒊𝒗𝒆𝒔𝑰𝒏𝑼𝑺𝑨 𝑿 ← 𝒃𝒐𝒓𝒏𝑰𝒏𝑼𝑺𝑨(𝑿) NormalAb-normal
  • 18. • Partial materialization bornInUSA livesInUSA stateless emigrant singer poet p1 ✓ ✓ p2 ✓ ✓ p3 ✓ ✓ p4 ✓ ✓ ✓ ✓ p5 ✓ ✓ ✓ ✓ p6 ✓ ✓ ✓ p7 ✓ ✓ ✓ p8 ✓ ✓ ✓ ✓ p9 ✓ ✓ ✓ ✓ ✓ p10 ✓ ✓ ✓ ✓ p11 ✓ ✓ ✓ ✓ 𝒓𝒊: 𝒍𝒊𝒗𝒆𝒔𝑰𝒏𝑼𝑺𝑨 𝑿 ← 𝒃𝒐𝒓𝒏𝑰𝒏𝑼𝑺𝑨(𝑿) NormalAb-normalRanking Rule’s Revisions 18
  • 19. Ranking Rule’s Revisions • Ordered partial materialization ranker • Only rules with higher quality • Ordered weighted partial materialization ranker • KG fact weight = 1 • Predicted facts inherit their weights from the rules 19
  • 20. Experiments • Ruleset quality 20 *Higher is better 0.60 0.65 0.70 0.75 0.80 20 40 60 80 100 Avg.Confidence Top-K (%) Rules YAGO3 Horn Naive Ordered & Weighted PM 0.8 0.82 0.84 0.86 0.88 0.9 0.92 0.94 20 40 60 80 100Avg.Confidence Top-K (%) Rules IMDB Horn Naive Ordered & Weighted PM Facts 10M 2M Rules 10K 25K
  • 21. Experiments • Predictions consistency 21*Lower is better 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 200 400 600 800 1000 ConflictRatio Top-K Rules YAGO3 Naive Ordered & Weighted PM 0 0.1 0.2 0.3 0.4 0.5 200 400 600 800 1000ConflictRatio Top-K Rules IMDB Naive Ordered & Weighted PM
  • 22. Experiments • Examples 22 𝑖𝑠𝑀𝑜𝑢𝑛𝑡𝑎𝑖𝑛(𝑋) ← 𝑖𝑠𝐼𝑛𝐴𝑢𝑠𝑡𝑟𝑖𝑎(𝑋), 𝑖𝑠𝐼𝑛𝐼𝑡𝑎𝑙𝑦 𝑋 , 𝑛𝑜𝑡 𝑖𝑠𝑅𝑖𝑣𝑒𝑟 (𝑋) 𝑖𝑠𝑃𝑜𝑙𝑖𝑡𝑂𝑓𝑈𝑆𝐴 𝑋 ← 𝑏𝑜𝑟𝑛𝐼𝑛𝑈𝑆𝐴 𝑋 , 𝑖𝑠𝐺𝑜𝑣 𝑋 , 𝑛𝑜𝑡 𝑖𝑠𝑃𝑜𝑙𝑖𝑡𝑃𝑢𝑒𝑟𝑡𝑜𝑅𝑖𝑐𝑜(𝑋) 𝑏𝑜𝑟𝑛𝐼𝑛𝑈𝑆𝐴 𝑋 ← 𝑎𝑐𝑡𝑒𝑑𝐼𝑛𝑀𝑜𝑣𝑖𝑒 𝑋 , 𝑐𝑟𝑒𝑎𝑡𝑒𝑑𝑀𝑜𝑣𝑖𝑒 𝑋 , 𝑛𝑜𝑡 𝑤𝑜𝑛𝐹𝑖𝑙𝑚𝑓𝑎𝑟𝑒(𝑋)
  • 23. Summary • Conclusion • Quality-based theory revision under OWA • Partial materialization for ranking revisions • Comparison of ranking methods on real life KGs • Outlook • Extending to higher arity predicates • Binary predicates [Tran et al., to appear ILP2016] • Evidence from text corpora • Exploiting partial completeness 23
  • 24. References • [Angiulli and Fassetti, 2014] Fabrizio Angiulli and Fabio Fassetti. Exploiting domain knowledge to detect outliers. Data Min. Knowl. Discov., 28(2):519–568, 2014. • [Dimopoulos and Kakas, 1995] Yannis Dimopoulos and Antonis C. Kakas. Learning non-monotonic logic programs: Learning exceptions. In Machine Learning: ECML-95, 8th European Conference on Machine Learning, Heraclion, Crete, Greece, April 25-27, 1995, Proceedings, pages 122–137, 1995. • [Galarraga et al., 2015] Luis Galarraga, Christina Teflioudi, Katja Hose, and Fabian M. Suchanek. Fast Rule Mining in Ontological Knowledge Bases with AMIE+. In VLDB Journal, 2015. • [Law et al., 2015] Mark Law, Alessandra Russo, and Krysia Broda. The ILASP system for learning answer set programs, 2015. • [Leone et al., 2006] Nicola Leone, Gerald Pfeifer, Wolfgang Faber, Thomas Eiter, Georg Gottlob, Simona Perri, and Francesco Scarcello. 2006. The DLV system for knowledge representation and reasoning. ACM Trans. Comput. Logic 7, 3 (July 2006), 499-562. • [Suzuki, 2006] Einoshin Suzuki. Data mining methods for discovering interesting exceptions from an unsupervised table. J. UCS, 12(6):627–653, 2006. • [Tran et al., 2016] Hai Dang Tran, Daria Stepanova, Mohamed H. Gad-Elrab, Francesca A. Lisi, Gerhard Weikum. Towards Nonmonotonic Relational Learning from Knowledge Graphs. ILP2016, London, UK, to appear. • [Katzouris et al., 2015] Nikos Katzouris, Alexander Artikis, and Georgios Paliouras. Incremental learning of event definitions with inductive logic programming. Machine Learning, 100(2-3):555– 585, 2015. 24
  • 25. Related Work • Learning nonmonotonic programs • E.g., [Dimopoulos and Kakas, 1995], ILASP [Law et al., 2015], ILED [Katzouris et al., 2015], etc. • Outlier detection in logic programs • E.g., [Angiulli and Fassetti, 2014], etc. • Mining exception rules • E.g., [Suzuki, 2006], etc. 25
  • 26. Problem Statement: Ruleset Quality • Independent Rule Measure (𝑟𝑚) • Support: 𝑠𝑢𝑝𝑝 𝐻 ← 𝐵 = 𝑠𝑢𝑝𝑝(𝐻 ∪ 𝐵) • Coverage: c𝑜𝑣 𝐻 ← 𝐵 = 𝑠𝑢𝑝𝑝(𝐵) • Confidence: c𝑜𝑛𝑓 𝐻 ← 𝐵 = 𝑠𝑢𝑝𝑝(𝐻∪𝐵) 𝑠𝑢𝑝𝑝(𝐵) • Lift: 𝑙𝑖𝑓𝑡 𝐻 ← 𝐵 = 𝑐𝑜𝑛𝑓(𝐻←𝐵) 𝑠𝑢𝑝𝑝(𝐻) • … • Average Ruleset Quality 26 𝑞 𝑟𝑚 𝑅 𝑁𝑀, 𝐺 = 𝑟 ∈𝑅 𝑁𝑀 𝑟𝑚(𝑟, 𝐺) 𝑅 𝑁𝑀
  • 27. Problem Statement: Conflicting Predictions • Measuring conflicts (auxiliary rules) 27 𝑟2: 𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑛𝑜𝑡 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋) 𝑟2 𝑎𝑢𝑥 : 𝑛𝑜𝑡_𝑟𝑒𝑙𝑒𝑎𝑠𝑒𝑑𝐼𝑛𝐽𝑃 𝑋 ← ℎ𝑎𝑠𝐽𝑃𝐶𝑜𝑚𝑝𝑜𝑠𝑒𝑟 𝑋 , 𝑝𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝑈𝑆𝐴(𝑋) 𝑞 𝑐𝑜𝑛𝑓𝑙𝑖𝑐𝑡 𝑅 𝑁𝑀, 𝐺 = |{(𝑝 𝑎 , 𝑛𝑜𝑡_𝑝 𝑎 ), … }| {𝑛𝑜𝑡_𝑝 𝑎 , … }
  • 28. Propositionalization • Unary predicates 28 Binary Predicates • hasType(enstein, scientist) • isMarriedTo(elsa, einstein) • bornIn(einstein, um) Unary • isAScientist(einstein) • isMarriedToEinstein(elsa) • bornInUlm(einstein) Abstraction • isAScientist(einstein) • isMarriedToScientist(elsa) • bornInGermany(einstein)
  • 29. Experiments • Input • Experiment statistics 29 YAGO3 IMDB Input Facts 10M 2M Horn Rules 10K 25K Revised Rules 6K 22K General-purpose KG Domain-specific KG (Movies)
  • 30. Experiments • Predictions assessment • Run DLV on YAGO and 𝑅 𝐻 then 𝑅 𝑁𝑀 seperately • Sample facts such that fact 𝑓 ∈ 𝑌𝐴𝐺𝑂 𝐻𝑌𝐴𝐺𝑂 𝑁𝑀 • 73% of the sampled facts were found to be erroneous 30 checked predictions
  • 31. Ranking Rule’s Revisions • Partial materialization ranker • Augment the original 𝐾𝐺 with predictions of other rules • Rank revisions on Avg. confidence of the 𝑟 and 𝑟 𝑎𝑢𝑥 31 𝑠𝑐𝑜𝑟𝑒 𝑟𝑒, 𝐾𝐺∗ = 𝑐𝑜𝑛𝑓 𝑟𝑒, 𝐾𝐺∗ + 𝑐𝑜𝑛𝑓(𝑟𝑒 𝑎𝑢𝑥, 𝐾𝐺∗) 2 where 𝑟𝑒 is the rule 𝑟 with exception 𝑒 & 𝐾𝐺∗ is the augmented 𝐾𝐺.