SlideShare a Scribd company logo

An Evolutionary Perspective on Approximate RDF Query Answering

Christophe Guéret, profile picture
Christophe Guéret

The document discusses the challenges of querying RDF data on the web, characterized by inconsistency, uncertainty, and rapid growth. It presents an evolutionary approach to approximate RDF query answering, focusing on generating solutions through evolutionary algorithms and verifying them using Bloom filters. The method aims to provide almost valid answers even when queries are unsatisfiable, enhancing the querying process by allowing the return of approximate data subsets.

TechnologyBusiness
1 of 73
Download to read offline
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
An Evolutionary Perspective on Approximate RDF Query Answering Christophe Guéret, Eyal Oren, Stefan Schlobach, Frank van Harmelen and Martijn Schut Vrije Universiteit, Amsterdam
Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF griffioen SUM 2008 - October 2, 2008 2 / 24
Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web griffioen SUM 2008 - October 2, 2008 2 / 24
Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! griffioen SUM 2008 - October 2, 2008 2 / 24
Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering griffioen SUM 2008 - October 2, 2008 2 / 24
Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion griffioen SUM 2008 - October 2, 2008 2 / 24
Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable griffioen SUM 2008 - October 2, 2008 2 / 24
Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable Approximate RDF Query answering griffioen SUM 2008 - October 2, 2008 2 / 24
Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable Approximate RDF Query answering Finding some, almost valid, data griffioen SUM 2008 - October 2, 2008 2 / 24
Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable Approximate RDF Query answering Finding some, almost valid, data The Evolutionary Perspective griffioen SUM 2008 - October 2, 2008 2 / 24
Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable Approximate RDF Query answering Finding some, almost valid, data The Evolutionary Perspective Test different solutions griffioen SUM 2008 - October 2, 2008 2 / 24
Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable Approximate RDF Query answering Finding some, almost valid, data The Evolutionary Perspective Test different solutions Progressive optimisation of the result griffioen SUM 2008 - October 2, 2008 2 / 24
Problem and context Method proposed Experimental results Conclusion 1 What’s the problem ? Querying RDF datastores Standard techniques 2 And Now for Something Completely Different Guessing the solution instead The way we do it 3 Does it work ? Evolution of the quality Some characteristics of this method 4 TODO list griffioen SUM 2008 - October 2, 2008 3 / 24
Problem and context Method proposed Experimental results Conclusion 1 What’s the problem ? Querying RDF datastores Standard techniques 2 And Now for Something Completely Different Guessing the solution instead The way we do it 3 Does it work ? Evolution of the quality Some characteristics of this method 4 TODO list griffioen SUM 2008 - October 2, 2008 4 / 24
Problem and context Method proposed Experimental results Conclusion Example RDF dataset <Ullman88> type Book . <Ullman88> label "Principles of Database and Knowledge-Base Systems" . <Ullman88> author b1 . b1 _1 ullman . ullman homepage <http://stanford.edu/~ullman/> . SPARQL query SELECT ?title WHERE { ?publication type Book . ?publication label ?title . } Expected answer griffioen ?title = "Principles of Database and Knowledge-Base Systems SUM 2008 - October 2, 2008 5 / 24
Problem and context Method proposed Experimental results Conclusion Problem description Triple = subject, predicate, object Dataset = graph of triples Querying : find a pattern in the graph griffioen A query and a graph [PSPARQL07] SUM 2008 - October 2, 2008 6 / 24
Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : griffioen SUM 2008 - October 2, 2008 7 / 24
Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book griffioen SUM 2008 - October 2, 2008 7 / 24
Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> griffioen SUM 2008 - October 2, 2008 7 / 24
Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> 2 Find all the possible results for ?publication label ?title griffioen SUM 2008 - October 2, 2008 7 / 24
Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> 2 Find all the possible results for ?publication label ?title ?publication ?title <Ullman88> "Principles of ..." griffioen SUM 2008 - October 2, 2008 7 / 24
Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> 2 Find all the possible results for ?publication label ?title ?publication ?title <Ullman88> "Principles of ..." 3 Do a join on the two tables and return the result griffioen SUM 2008 - October 2, 2008 7 / 24
Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> 2 Find all the possible results for ?publication label ?title ?publication ?title <Ullman88> "Principles of ..." 3 Do a join on the two tables and return the result ?title = "Principles of ..." griffioen SUM 2008 - October 2, 2008 7 / 24
Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> 2 Find all the possible results for ?publication label ?title ?publication ?title <Ullman88> "Principles of ..." 3 Do a join on the two tables and return the result ?title = "Principles of ..." Fast thanks to the creation of indexes and query optimisation griffioen SUM 2008 - October 2, 2008 7 / 24
Problem and context Method proposed Experimental results Conclusion Motivation Designed to return results only when there are some Not designed for incomplete and approximate queries/answers Hard to distribute griffioen SUM 2008 - October 2, 2008 8 / 24
Problem and context Method proposed Experimental results Conclusion Motivation Designed to return results only when there are some Not designed for incomplete and approximate queries/answers Hard to distribute Approximate answers to precise queries If the query is unsat, return the best almost sat solution found griffioen SUM 2008 - October 2, 2008 8 / 24
Problem and context Method proposed Experimental results Conclusion Motivation Designed to return results only when there are some Not designed for incomplete and approximate queries/answers Hard to distribute Approximate answers to precise queries If the query is unsat, return the best almost sat solution found Precises answers to approximate queries Return a subset of existing solutions instead of showing them all griffioen SUM 2008 - October 2, 2008 8 / 24
Problem and context Method proposed Experimental results Conclusion Motivation Designed to return results only when there are some Not designed for incomplete and approximate queries/answers Hard to distribute Approximate answers to precise queries If the query is unsat, return the best almost sat solution found Precises answers to approximate queries Return a subset of existing solutions instead of showing them all Interactive querying Use of intermediate results to help the user improving his griffioen query SUM 2008 - October 2, 2008 8 / 24
Problem and context Method proposed Experimental results Conclusion 1 What’s the problem ? Querying RDF datastores Standard techniques 2 And Now for Something Completely Different Guessing the solution instead The way we do it 3 Does it work ? Evolution of the quality Some characteristics of this method 4 TODO list griffioen SUM 2008 - October 2, 2008 9 / 24
Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : griffioen SUM 2008 - October 2, 2008 10 / 24
Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables griffioen SUM 2008 - October 2, 2008 10 / 24
Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book griffioen SUM 2008 - October 2, 2008 10 / 24
Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid griffioen SUM 2008 - October 2, 2008 10 / 24
Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid Triple Is in the graph ? <Ullman88> type Book yes <Ullman88> label Book no griffioen SUM 2008 - October 2, 2008 10 / 24
Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid Triple Is in the graph ? <Ullman88> type Book yes <Ullman88> label Book no 3 If the solution is OK, stop. Otherwise, try again with something else griffioen SUM 2008 - October 2, 2008 10 / 24
Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid Triple Is in the graph ? <Ullman88> type Book yes <Ullman88> label Book no 3 If the solution is OK, stop. Otherwise, try again with something else Rely on membership testing (instead of lookup) griffioen SUM 2008 - October 2, 2008 10 / 24
Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid Triple Is in the graph ? <Ullman88> type Book yes <Ullman88> label Book no 3 If the solution is OK, stop. Otherwise, try again with something else Rely on membership testing (instead of lookup) The testing loop can be stopped at any time griffioen SUM 2008 - October 2, 2008 10 / 24
Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid Triple Is in the graph ? <Ullman88> type Book yes <Ullman88> label Book no 3 If the solution is OK, stop. Otherwise, try again with something else Rely on membership testing (instead of lookup) The testing loop can be stopped at any time A result may satisfy part of the query griffioen SUM 2008 - October 2, 2008 10 / 24
Problem and context Method proposed Experimental results Conclusion Our choices Need to pay attention to two aspects griffioen SUM 2008 - October 2, 2008 11 / 24
Problem and context Method proposed Experimental results Conclusion Our choices Need to pay attention to two aspects 1 Each try should be a step closer to the solution Random guessing may never end Stopping the process at t + 1 should give better results than at t griffioen SUM 2008 - October 2, 2008 11 / 24
Problem and context Method proposed Experimental results Conclusion Our choices Need to pay attention to two aspects 1 Each try should be a step closer to the solution Random guessing may never end Stopping the process at t + 1 should give better results than at t 2 Testing a candidate solution must be fast Will try a lot of solutions griffioen SUM 2008 - October 2, 2008 11 / 24
Problem and context Method proposed Experimental results Conclusion Our choices Need to pay attention to two aspects 1 Each try should be a step closer to the solution Random guessing may never end Stopping the process at t + 1 should give better results than at t 2 Testing a candidate solution must be fast Will try a lot of solutions We made the following choices Generation of solutions : Evolutionary algorithm Verification of solutions : Bloom filter based testing griffioen SUM 2008 - October 2, 2008 11 / 24
Problem and context Method proposed Experimental results Conclusion Binary Bloom filters (1/2) Compact representation of information : a set of n = 8 bits 1 2 3 4 5 6 7 8 Supports two operations INSERT ( KEY ) : Insert a key into the filter CONTAINS ( KEY ) : Test for the presence of a key Use k = 3 hash functions to compute a set of bits from a key HASH 1(“ HELLO WORLD ”)=8 HASH 2(“ HELLO WORLD ”)=6 HASH 3(“ HELLO WORLD ”)=3 griffioen SUM 2008 - October 2, 2008 12 / 24
Problem and context Method proposed Experimental results Conclusion Binary Bloom filters (2/2) INSERT (“ HELLO WORLD ”) Current Bit-wise or operation OR Always successful (i.e. “Hello world” unlimited capacity) = Precision depends of New number of elements m. CONTAINS (“B ONJOUR !”) Current Bit-wise and operation AND “Bonjour !” Positive result can be a collision = kn griffioen Test result perror = (1 − e− m )k SUM 2008 - October 2, 2008 13 / 24
Problem and context Method proposed Experimental results Conclusion A first (naive) approach Insert all the triples into a unique Bloom filter. INSERT (“<Ullman88>_type_Book”) INSERT (“<Ullman88>_label_"Principles of ..."”) ... griffioen SUM 2008 - October 2, 2008 14 / 24
Problem and context Method proposed Experimental results Conclusion A first (naive) approach Insert all the triples into a unique Bloom filter. INSERT (“<Ullman88>_type_Book”) INSERT (“<Ullman88>_label_"Principles of ..."”) ... Use the CONTAINS operation to verify a solution CONTAINS (“<Ullman88>_type_Book”) ⇒ true CONTAINS (“<Ullman88>_label_Book”) ⇒ false griffioen SUM 2008 - October 2, 2008 14 / 24
Problem and context Method proposed Experimental results Conclusion A first (naive) approach Insert all the triples into a unique Bloom filter. INSERT (“<Ullman88>_type_Book”) INSERT (“<Ullman88>_label_"Principles of ..."”) ... Use the CONTAINS operation to verify a solution CONTAINS (“<Ullman88>_type_Book”) ⇒ true CONTAINS (“<Ullman88>_label_Book”) ⇒ false Not the best approach ! Let’s see what happen in detail . . . griffioen SUM 2008 - October 2, 2008 14 / 24
Problem and context Method proposed Experimental results Conclusion A first (naive) approach Insert all the triples into a unique Bloom filter. INSERT (“<Ullman88>_type_Book”) INSERT (“<Ullman88>_label_"Principles of ..."”) ... Use the CONTAINS operation to verify a solution CONTAINS (“<Ullman88>_type_Book”) ⇒ true CONTAINS (“<Ullman88>_label_Book”) ⇒ false Not the best approach ! Let’s see what happen in detail . . . ?publication label ?title CONTAINS (“<Ullman88>_label_Book”) griffioen modify ?publication and ?title SUM 2008 - October 2, 2008 14 / 24
Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book griffioen SUM 2008 - October 2, 2008 15 / 24
Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book SPO griffioen SUM 2008 - October 2, 2008 15 / 24
Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book <Ullman88>_type SPO SP griffioen SUM 2008 - October 2, 2008 15 / 24
Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book <Ullman88>_type type_Book SPO SP PO griffioen SUM 2008 - October 2, 2008 15 / 24
Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book <Ullman88>_type type_Book <Ullman88>_Boo SPO SP PO SO griffioen SUM 2008 - October 2, 2008 15 / 24
Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book <Ullman88>_type type_Book <Ullman88>_Boo SPO SP PO SO Three domains are defined S = <Ullman88> b1 ullman P = type label author _1 homepage O = Book "Principles of ..." b1 ullman <http://...> griffioen SUM 2008 - October 2, 2008 15 / 24
Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book <Ullman88>_type type_Book <Ullman88>_Boo SPO SP PO SO Three domains are defined S = <Ullman88> b1 ullman P = type label author _1 homepage O = Book "Principles of ..." b1 ullman <http://...> Each term is replaced by an integer (with a dictionary) griffioen <Ullman88> → 46 SUM 2008 - October 2, 2008 15 / 24
Problem and context Method proposed Experimental results Conclusion Evolutionary algorithm flowchart [Eiben2003] Set of populations + Set of operators griffioen SUM 2008 - October 2, 2008 16 / 24
Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title griffioen SUM 2008 - October 2, 2008 17 / 24
Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title Creation of constraints to verify griffioen SUM 2008 - October 2, 2008 17 / 24
Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title Creation of constraints to verify Clause ?publication type Book . bloom(spo |?publication1 type Book) bloom(sp |?publication1 type) bloom(po |type Book) griffioen SUM 2008 - October 2, 2008 17 / 24
Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title Creation of constraints to verify Clause ?publication type Book . bloom(spo |?publication1 type Book) bloom(sp |?publication1 type) bloom(po |type Book) Clause ?publication label ?title . bloom(spo |?publication2 label ?title) bloom(sp |?publication2 label) bloom(po |label ?title) bloom(so |?publication2 ?title) griffioen SUM 2008 - October 2, 2008 17 / 24
Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title Creation of constraints to verify Clause ?publication type Book . bloom(spo |?publication1 type Book) bloom(sp |?publication1 type) bloom(po |type Book) Clause ?publication label ?title . bloom(spo |?publication2 label ?title) bloom(sp |?publication2 label) bloom(po |label ?title) bloom(so |?publication2 ?title) Equality constraint equal(?publication1 ,?publication2 ) griffioen SUM 2008 - October 2, 2008 17 / 24
Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title Removed Creation of constraints to verify because Clause ?publication type Book . always true bloom(spo |?publication type Book) 1 bloom(sp |?publication1 type) bloom(po |type Book) Clause ?publication label ?title . bloom(spo |?publication2 label ?title) bloom(sp |?publication2 label) bloom(po |label ?title) bloom(so |?publication2 ?title) Equality constraint equal(?publication1 ,?publication2 ) griffioen SUM 2008 - October 2, 2008 17 / 24
Problem and context Method proposed Experimental results Conclusion Evaluation of a candidate solution Solution is checked against all the constraints. If one is satisfied, A global reward w is won Each variable used is equally rewarded Rewards for : bloom(spo|?publication2 label ?title) reward(solution) += w w reward(?publication1 ) += 2 reward(?title) += w 2 griffioen SUM 2008 - October 2, 2008 18 / 24
Problem and context Method proposed Experimental results Conclusion Creation of new individuals Select two individuals and do a one point crossover dblp:ullman <Ullman88> "Principles. . . " dblp:ullman <Ullman88> _:b1 <Ullman88> dblp:ullman _:b1 <Ullman88> dblp:ullman "Principles. . . " Randomly pick a pivot point Swap the two parts Mutate the least efficient variable dblp:ullman <Ullman88> "Principles. . . " 0 3×w 2×w <Ullman88> <Ullman88> "Principles. . . " Select the variable with lowest Assign a random new value reward griffioen SUM 2008 - October 2, 2008 19 / 24
Problem and context Method proposed Experimental results Conclusion 1 What’s the problem ? Querying RDF datastores Standard techniques 2 And Now for Something Completely Different Guessing the solution instead The way we do it 3 Does it work ? Evolution of the quality Some characteristics of this method 4 TODO list griffioen SUM 2008 - October 2, 2008 20 / 24
Problem and context Method proposed Experimental results Conclusion Results on some (small) datasets Database FOAF (15k triples) and DBLP (3M triples) Query with, respectively, 4 and 11 different variables Average result for 200 individuals and 500 generations 60 100 50 90 fitness value fitness value 40 30 80 20 70 10 0 60 0 100 200 300 400 500 0 100 200 300 400 500 n-th generation n-th generation Solutions with maximum reward (52) are found for FOAF griffioen Not enough time for DBLP (max 319) SUM 2008 - October 2, 2008 21 / 24
Problem and context Method proposed Experimental results Conclusion Scalibility & speed Low memory requirements Only depends on the number of individuals and the size of the Bloom filters (a) parsing (b) querying dataset memory dataset memory FOAF 65 MB FOAF 15 MB DBLP 230 MB DBLP 140 MB Table: Average memory usage (mostly due to dictionary) Computation can be distributed Candidate solutions are independent The dictionary can be based on a DHT griffioen SUM 2008 - October 2, 2008 22 / 24
Problem and context Method proposed Experimental results Conclusion 1 What’s the problem ? Querying RDF datastores Standard techniques 2 And Now for Something Completely Different Guessing the solution instead The way we do it 3 Does it work ? Evolution of the quality Some characteristics of this method 4 TODO list griffioen SUM 2008 - October 2, 2008 23 / 24
Problem and context Method proposed Experimental results Conclusion Status and future work Current status The search process can be slow to converge Several parameters to tune (rewards, size of the population, number of generations, . . . ) griffioen SUM 2008 - October 2, 2008 24 / 24
Problem and context Method proposed Experimental results Conclusion Status and future work Current status The search process can be slow to converge Several parameters to tune (rewards, size of the population, number of generations, . . . ) Current work griffioen SUM 2008 - October 2, 2008 24 / 24
Problem and context Method proposed Experimental results Conclusion Status and future work Current status The search process can be slow to converge Several parameters to tune (rewards, size of the population, number of generations, . . . ) Current work 1 Improve benchmarking Test with more queries and more datasets Better study of the influence of the parameters griffioen SUM 2008 - October 2, 2008 24 / 24
Problem and context Method proposed Experimental results Conclusion Status and future work Current status The search process can be slow to converge Several parameters to tune (rewards, size of the population, number of generations, . . . ) Current work 1 Improve benchmarking Test with more queries and more datasets Better study of the influence of the parameters 2 Improve evolution Experiment different type of crossover and mutation Implement dynamic valuations for the rewards Improve early results on tabbu search approach griffioen SUM 2008 - October 2, 2008 24 / 24
Problem and context Method proposed Experimental results Conclusion Status and future work Current status The search process can be slow to converge Several parameters to tune (rewards, size of the population, number of generations, . . . ) Current work 1 Improve benchmarking Test with more queries and more datasets Better study of the influence of the parameters 2 Improve evolution Experiment different type of crossover and mutation Implement dynamic valuations for the rewards Improve early results on tabbu search approach 3 Test other, easy to parallelize and anytime, optimizer Swarm based algorithm (PSO, ...) or an other EA griffioen CSP solver SUM 2008 - October 2, 2008 24 / 24

More Related Content

PDF
Question Answering - Application and Challenges
Jens Lehmann
 
PPTX
Question answering
Nafiseh Navabpour
 
PPTX
From TREC to Watson: is open domain question answering a solved problem?
Constantin Orasan
 
PPTX
R programming for psychometrics
Diane Talley
 
PDF
Open domain Question Answering System - Research project in NLP
GVS Chaitanya
 
PDF
Practical machine learning - Part 1
Traian Rebedea
 
PDF
Natural language processing for requirements engineering: ICSE 2021 Technical...
alessio_ferrari
 
PPTX
Using a keyword extraction pipeline to understand concepts in future work sec...
Kai Li
 
Question Answering - Application and Challenges
Jens Lehmann
 
Question answering
Nafiseh Navabpour
 
From TREC to Watson: is open domain question answering a solved problem?
Constantin Orasan
 
R programming for psychometrics
Diane Talley
 
Open domain Question Answering System - Research project in NLP
GVS Chaitanya
 
Practical machine learning - Part 1
Traian Rebedea
 
Natural language processing for requirements engineering: ICSE 2021 Technical...
alessio_ferrari
 
Using a keyword extraction pipeline to understand concepts in future work sec...
Kai Li
 

What's hot (8)

PDF
Combining IR with Relevance Feedback for Concept Location
Sonia Haiduc
 
PPTX
Automatic Key Term Extraction from Spoken Course Lectures
Yun-Nung (Vivian) Chen
 
PPTX
Automatic Key Term Extraction and Summarization from Spoken Course Lectures
Yun-Nung (Vivian) Chen
 
PDF
Extracting keywords from texts - Sanda Martincic Ipsic
Institute of Contemporary Sciences
 
PPTX
Setting Up a Qualitative or Mixed Methods Research Project in NVivo 10 to Cod...
Shalin Hai-Jew
 
PDF
2011 EASE - Motivation in Software Engineering: A Systematic Review Update
HASE – Human Aspects in Software Engineering
 
PPT
QALL-ME: Ontology and Semantic Web
Constantin Orasan
 
PDF
Arabic Question Answering: Challenges, Tasks, Approaches, Test-sets, Tools, A...
Ahmed Magdy Ezzeldin, MSc.
 
Combining IR with Relevance Feedback for Concept Location
Sonia Haiduc
 
Automatic Key Term Extraction from Spoken Course Lectures
Yun-Nung (Vivian) Chen
 
Automatic Key Term Extraction and Summarization from Spoken Course Lectures
Yun-Nung (Vivian) Chen
 
Extracting keywords from texts - Sanda Martincic Ipsic
Institute of Contemporary Sciences
 
Setting Up a Qualitative or Mixed Methods Research Project in NVivo 10 to Cod...
Shalin Hai-Jew
 
2011 EASE - Motivation in Software Engineering: A Systematic Review Update
HASE – Human Aspects in Software Engineering
 
QALL-ME: Ontology and Semantic Web
Constantin Orasan
 
Arabic Question Answering: Challenges, Tasks, Approaches, Test-sets, Tools, A...
Ahmed Magdy Ezzeldin, MSc.
 
Ad

Viewers also liked (20)

PPT
Drugan Notes- Biological Perspective
Kim Drugan
 
PPT
Ch. 2 -_the_biological_perspective
mcolon344
 
PDF
Every Crisis is Global, Social, Viral
Gaurav Mishra
 
PPTX
Learning styles from a multicultural perspective
Namchalla LSS
 
PDF
2010 1 materialism1
Jonathan Dunnemann
 
PPT
Attitude Changes Everything
pudding37
 
PPTX
The bee effect: Action to effect change
Allen McClinton
 
PPTX
Valuing ecosystem services: a biological perspective
Kent Holsinger
 
PDF
Managing Multicultural Individuals
Margareta Heidt
 
PPTX
Inculcate Self Confidence & Self Belief
Health Education Library for People
 
PPTX
Why Watson Won: A cognitive perspective
James Hendler
 
PPTX
C:\Multicultural Education Powerpoint
nkiogima
 
PPT
Evolutionary perspective
mpape
 
PPTX
our behaviour is the foundation of our attitude and self perception
Parveen Bano
 
PPTX
Theorizing the Future of Computer-Mediated Communication: The Changing Role o...
Jessica Vitak
 
PPTX
Cognition, Learning, and Self-Tracking - Quantified Self 2011
nickwinter
 
DOC
ATTITUDE AND BEHAVIOUR
ANTHONY ALU
 
PPT
Meth Powerpoint
ErinGullberg
 
PPTX
Social cognition
Col Mukteshwar Prasad
 
PDF
Social cognition
Surashmie Kaalmegh
 
Drugan Notes- Biological Perspective
Kim Drugan
 
Ch. 2 -_the_biological_perspective
mcolon344
 
Every Crisis is Global, Social, Viral
Gaurav Mishra
 
Learning styles from a multicultural perspective
Namchalla LSS
 
2010 1 materialism1
Jonathan Dunnemann
 
Attitude Changes Everything
pudding37
 
The bee effect: Action to effect change
Allen McClinton
 
Valuing ecosystem services: a biological perspective
Kent Holsinger
 
Managing Multicultural Individuals
Margareta Heidt
 
Inculcate Self Confidence & Self Belief
Health Education Library for People
 
Why Watson Won: A cognitive perspective
James Hendler
 
C:\Multicultural Education Powerpoint
nkiogima
 
Evolutionary perspective
mpape
 
our behaviour is the foundation of our attitude and self perception
Parveen Bano
 
Theorizing the Future of Computer-Mediated Communication: The Changing Role o...
Jessica Vitak
 
Cognition, Learning, and Self-Tracking - Quantified Self 2011
nickwinter
 
ATTITUDE AND BEHAVIOUR
ANTHONY ALU
 
Meth Powerpoint
ErinGullberg
 
Social cognition
Col Mukteshwar Prasad
 
Social cognition
Surashmie Kaalmegh
 
Ad

Similar to An Evolutionary Perspective on Approximate RDF Query Answering (20)

PPTX
Logical Detection of Invalid SameAs Statements in RDF Data
University PARIS-SUD
 
PPTX
20130622 okfn hackathon t2
Seonho Kim
 
PDF
Modelling context and statement-level metadata in knowledge graphs
Fabrizio Orlandi
 
PPT
Ks2007 Semanticweb In Action
Rinke Hoekstra
 
PPTX
‘Facilitating User Engagement by Enriching Library Data using Semantic Techno...
CONUL Conference
 
PPTX
Validating statistical Index Data represented in RDF using SPARQL Queries: Co...
Jose Emilio Labra Gayo
 
PDF
Hide the Stack: Toward Usable Linked Data
aba-sah
 
PDF
Uplift – Generating RDF datasets from non-RDF data with R2RML
Christophe Debruyne
 
PPTX
Mining and Managing Large-scale Linked Open Data
MOVING Project
 
PPTX
Mining and Managing Large-scale Linked Open Data
Ansgar Scherp
 
PPTX
CBS CEDAR Presentation
Albert Meroño-Peñuela
 
PPTX
RDF2Rule PRESENTATION
Efrah Shakir
 
PDF
QUALITY-AWARE SUBGRAPH MATCHING OVER INCONSISTENT PROBABILISTIC GRAPH DATABASES
Nexgen Technology
 
PPTX
Representing verifiable statistical index computations as linked data
Jose Emilio Labra Gayo
 
PDF
Which Model Does Not Belong: A Dialogue
Bertram Ludäscher
 
PPT
R for the semantic web, Quesada useR 2009
Jose Quesada
 
PDF
Using Page Size for Controlling Duplicate Query Results in Semantic Web
IJwest
 
PPTX
How the Web can change social science research (including yours)
Frank van Harmelen
 
PPTX
Querying the Web of Data
Rinke Hoekstra
 
PDF
Interactive Knowledge Discovery over Web of Data.
Mehwish Alam
 
Logical Detection of Invalid SameAs Statements in RDF Data
University PARIS-SUD
 
20130622 okfn hackathon t2
Seonho Kim
 
Modelling context and statement-level metadata in knowledge graphs
Fabrizio Orlandi
 
Ks2007 Semanticweb In Action
Rinke Hoekstra
 
‘Facilitating User Engagement by Enriching Library Data using Semantic Techno...
CONUL Conference
 
Validating statistical Index Data represented in RDF using SPARQL Queries: Co...
Jose Emilio Labra Gayo
 
Hide the Stack: Toward Usable Linked Data
aba-sah
 
Uplift – Generating RDF datasets from non-RDF data with R2RML
Christophe Debruyne
 
Mining and Managing Large-scale Linked Open Data
MOVING Project
 
Mining and Managing Large-scale Linked Open Data
Ansgar Scherp
 
CBS CEDAR Presentation
Albert Meroño-Peñuela
 
RDF2Rule PRESENTATION
Efrah Shakir
 
QUALITY-AWARE SUBGRAPH MATCHING OVER INCONSISTENT PROBABILISTIC GRAPH DATABASES
Nexgen Technology
 
Representing verifiable statistical index computations as linked data
Jose Emilio Labra Gayo
 
Which Model Does Not Belong: A Dialogue
Bertram Ludäscher
 
R for the semantic web, Quesada useR 2009
Jose Quesada
 
Using Page Size for Controlling Duplicate Query Results in Semantic Web
IJwest
 
How the Web can change social science research (including yours)
Frank van Harmelen
 
Querying the Web of Data
Rinke Hoekstra
 
Interactive Knowledge Discovery over Web of Data.
Mehwish Alam
 

More from Christophe Guéret (20)

PDF
HHAI June 2022 - KGs and Hybrid Intelligence
Christophe Guéret
 
PDF
Informal presentation about RES
Christophe Guéret
 
ODP
Stop making tools! Nobody likes them anyway...
Christophe Guéret
 
ODP
The Entity Registry System: Collaborative Editing of Entity Data in Poorly Co...
Christophe Guéret
 
ODP
Introduction about WorldWideSemanticWeb.org for the workshop "Making it Matter"
Christophe Guéret
 
PDF
The Entity Registry System (ERS)
Christophe Guéret
 
PDF
Let's downscale the semantic web !
Christophe Guéret
 
PDF
Your next data viz gear should be a Wii-U
Christophe Guéret
 
PDF
Linking knowledge spaces
Christophe Guéret
 
ODP
The data behind the HuisKluis
Christophe Guéret
 
PDF
Digital archiving 3.0
Christophe Guéret
 
PDF
The road towards a Web-based data ecosystem
Christophe Guéret
 
PDF
Linked Open Data for Digital Humanities
Christophe Guéret
 
PDF
Downscaling information systems for education
Christophe Guéret
 
PDF
ICT4D course 2013 - Low resources infrastructure
Christophe Guéret
 
PDF
ICT4D course 2013 - OLPC deployments
Christophe Guéret
 
PDF
ICT4D course 2013 - Sugar
Christophe Guéret
 
PDF
Exposing the data from NARCIS with VIVO
Christophe Guéret
 
PDF
Clarifier le sens de vos données publiques avec le Web de données
Christophe Guéret
 
ODP
Embedding young learners into the information society
Christophe Guéret
 
HHAI June 2022 - KGs and Hybrid Intelligence
Christophe Guéret
 
Informal presentation about RES
Christophe Guéret
 
Stop making tools! Nobody likes them anyway...
Christophe Guéret
 
The Entity Registry System: Collaborative Editing of Entity Data in Poorly Co...
Christophe Guéret
 
Introduction about WorldWideSemanticWeb.org for the workshop "Making it Matter"
Christophe Guéret
 
The Entity Registry System (ERS)
Christophe Guéret
 
Let's downscale the semantic web !
Christophe Guéret
 
Your next data viz gear should be a Wii-U
Christophe Guéret
 
Linking knowledge spaces
Christophe Guéret
 
The data behind the HuisKluis
Christophe Guéret
 
Digital archiving 3.0
Christophe Guéret
 
The road towards a Web-based data ecosystem
Christophe Guéret
 
Linked Open Data for Digital Humanities
Christophe Guéret
 
Downscaling information systems for education
Christophe Guéret
 
ICT4D course 2013 - Low resources infrastructure
Christophe Guéret
 
ICT4D course 2013 - OLPC deployments
Christophe Guéret
 
ICT4D course 2013 - Sugar
Christophe Guéret
 
Exposing the data from NARCIS with VIVO
Christophe Guéret
 
Clarifier le sens de vos données publiques avec le Web de données
Christophe Guéret
 
Embedding young learners into the information society
Christophe Guéret
 

Recently uploaded (20)

PDF
How UI/UX Design Impacts User Retention in Mobile Apps.pdf
SynapseIndia
 
PDF
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
Jian-Hong Pan
 
DOCX
The AUB Centre for AI in Media Proposal.docx
GAONE9
 
PDF
Bridging biosciences and deep learning for revolutionary discoveries: a compr...
IAESIJAI
 
PDF
The Rise and Fall of 3GPP – Time for a Sabbatical?
3G4G
 
PPTX
Understanding_Digital_Forensics_Presentation.pptx
ImranKhan423233
 
PPTX
Digital-Transformation-Roadmap-for-Companies.pptx
David Malick Dieng
 
PDF
Modernizing your data center with Dell and AMD
Principled Technologies
 
PDF
Encapsulation_ Review paper, used for researhc scholars
gurumoop
 
PDF
Approach and Philosophy of On baking technology
30anthuong17
 
PPT
“AI and Expert System Decision Support & Business Intelligence Systems”
ZubinRadhakrishnan
 
PPTX
20250228 LYD VKU AI Blended-Learning.pptx
DatVoNgoc
 
PPTX
VMware vSphere Foundation How to Sell Presentation-Ver1.4-2-14-2024.pptx
blackmambaettijean
 
PDF
Unlocking AI with Model Context Protocol (MCP)
Brian McKeiver
 
PDF
Peak of Data & AI Encore- AI for Metadata and Smarter Workflows
Safe Software
 
PPTX
PA Analog/Digital System: The Backbone of Modern Surveillance and Communication
AVTRON Technologies LLC
 
PPTX
MYSQL Presentation for SQL database connectivity
Swati270511
 
PDF
Electronic commerce courselecture one. Pdf
OmerMohamed64
 
PDF
Machine learning based COVID-19 study performance prediction
IAESIJAI
 
PDF
Architecting across the Boundaries of two Complex Domains - Healthcare & Tech...
Peter Tan
 
How UI/UX Design Impacts User Retention in Mobile Apps.pdf
SynapseIndia
 
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
Jian-Hong Pan
 
The AUB Centre for AI in Media Proposal.docx
GAONE9
 
Bridging biosciences and deep learning for revolutionary discoveries: a compr...
IAESIJAI
 
The Rise and Fall of 3GPP – Time for a Sabbatical?
3G4G
 
Understanding_Digital_Forensics_Presentation.pptx
ImranKhan423233
 
Digital-Transformation-Roadmap-for-Companies.pptx
David Malick Dieng
 
Modernizing your data center with Dell and AMD
Principled Technologies
 
Encapsulation_ Review paper, used for researhc scholars
gurumoop
 
Approach and Philosophy of On baking technology
30anthuong17
 
“AI and Expert System Decision Support & Business Intelligence Systems”
ZubinRadhakrishnan
 
20250228 LYD VKU AI Blended-Learning.pptx
DatVoNgoc
 
VMware vSphere Foundation How to Sell Presentation-Ver1.4-2-14-2024.pptx
blackmambaettijean
 
Unlocking AI with Model Context Protocol (MCP)
Brian McKeiver
 
Peak of Data & AI Encore- AI for Metadata and Smarter Workflows
Safe Software
 
PA Analog/Digital System: The Backbone of Modern Surveillance and Communication
AVTRON Technologies LLC
 
MYSQL Presentation for SQL database connectivity
Swati270511
 
Electronic commerce courselecture one. Pdf
OmerMohamed64
 
Machine learning based COVID-19 study performance prediction
IAESIJAI
 
Architecting across the Boundaries of two Complex Domains - Healthcare & Tech...
Peter Tan
 

An Evolutionary Perspective on Approximate RDF Query Answering

  • 1. An Evolutionary Perspective on Approximate RDF Query Answering Christophe Guéret, Eyal Oren, Stefan Schlobach, Frank van Harmelen and Martijn Schut Vrije Universiteit, Amsterdam
  • 2. Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF griffioen SUM 2008 - October 2, 2008 2 / 24
  • 3. Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web griffioen SUM 2008 - October 2, 2008 2 / 24
  • 4. Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! griffioen SUM 2008 - October 2, 2008 2 / 24
  • 5. Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering griffioen SUM 2008 - October 2, 2008 2 / 24
  • 6. Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion griffioen SUM 2008 - October 2, 2008 2 / 24
  • 7. Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable griffioen SUM 2008 - October 2, 2008 2 / 24
  • 8. Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable Approximate RDF Query answering griffioen SUM 2008 - October 2, 2008 2 / 24
  • 9. Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable Approximate RDF Query answering Finding some, almost valid, data griffioen SUM 2008 - October 2, 2008 2 / 24
  • 10. Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable Approximate RDF Query answering Finding some, almost valid, data The Evolutionary Perspective griffioen SUM 2008 - October 2, 2008 2 / 24
  • 11. Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable Approximate RDF Query answering Finding some, almost valid, data The Evolutionary Perspective Test different solutions griffioen SUM 2008 - October 2, 2008 2 / 24
  • 12. Problem and context Method proposed Experimental results Conclusion The next 30 minutes in 4 points... RDF Data on the Web Inconsistent, uncertain, heterogeneous, Huge and growing! RDF Query answering Finding data matching criterion ... many queries are actually not satisfiable Approximate RDF Query answering Finding some, almost valid, data The Evolutionary Perspective Test different solutions Progressive optimisation of the result griffioen SUM 2008 - October 2, 2008 2 / 24
  • 13. Problem and context Method proposed Experimental results Conclusion 1 What’s the problem ? Querying RDF datastores Standard techniques 2 And Now for Something Completely Different Guessing the solution instead The way we do it 3 Does it work ? Evolution of the quality Some characteristics of this method 4 TODO list griffioen SUM 2008 - October 2, 2008 3 / 24
  • 14. Problem and context Method proposed Experimental results Conclusion 1 What’s the problem ? Querying RDF datastores Standard techniques 2 And Now for Something Completely Different Guessing the solution instead The way we do it 3 Does it work ? Evolution of the quality Some characteristics of this method 4 TODO list griffioen SUM 2008 - October 2, 2008 4 / 24
  • 15. Problem and context Method proposed Experimental results Conclusion Example RDF dataset <Ullman88> type Book . <Ullman88> label "Principles of Database and Knowledge-Base Systems" . <Ullman88> author b1 . b1 _1 ullman . ullman homepage <http://stanford.edu/~ullman/> . SPARQL query SELECT ?title WHERE { ?publication type Book . ?publication label ?title . } Expected answer griffioen ?title = "Principles of Database and Knowledge-Base Systems SUM 2008 - October 2, 2008 5 / 24
  • 16. Problem and context Method proposed Experimental results Conclusion Problem description Triple = subject, predicate, object Dataset = graph of triples Querying : find a pattern in the graph griffioen A query and a graph [PSPARQL07] SUM 2008 - October 2, 2008 6 / 24
  • 17. Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : griffioen SUM 2008 - October 2, 2008 7 / 24
  • 18. Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book griffioen SUM 2008 - October 2, 2008 7 / 24
  • 19. Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> griffioen SUM 2008 - October 2, 2008 7 / 24
  • 20. Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> 2 Find all the possible results for ?publication label ?title griffioen SUM 2008 - October 2, 2008 7 / 24
  • 21. Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> 2 Find all the possible results for ?publication label ?title ?publication ?title <Ullman88> "Principles of ..." griffioen SUM 2008 - October 2, 2008 7 / 24
  • 22. Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> 2 Find all the possible results for ?publication label ?title ?publication ?title <Ullman88> "Principles of ..." 3 Do a join on the two tables and return the result griffioen SUM 2008 - October 2, 2008 7 / 24
  • 23. Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> 2 Find all the possible results for ?publication label ?title ?publication ?title <Ullman88> "Principles of ..." 3 Do a join on the two tables and return the result ?title = "Principles of ..." griffioen SUM 2008 - October 2, 2008 7 / 24
  • 24. Problem and context Method proposed Experimental results Conclusion Standard techniques Standard approach : 1 Find all the possible results for ?publication type Book ?publication <Ullman88> 2 Find all the possible results for ?publication label ?title ?publication ?title <Ullman88> "Principles of ..." 3 Do a join on the two tables and return the result ?title = "Principles of ..." Fast thanks to the creation of indexes and query optimisation griffioen SUM 2008 - October 2, 2008 7 / 24
  • 25. Problem and context Method proposed Experimental results Conclusion Motivation Designed to return results only when there are some Not designed for incomplete and approximate queries/answers Hard to distribute griffioen SUM 2008 - October 2, 2008 8 / 24
  • 26. Problem and context Method proposed Experimental results Conclusion Motivation Designed to return results only when there are some Not designed for incomplete and approximate queries/answers Hard to distribute Approximate answers to precise queries If the query is unsat, return the best almost sat solution found griffioen SUM 2008 - October 2, 2008 8 / 24
  • 27. Problem and context Method proposed Experimental results Conclusion Motivation Designed to return results only when there are some Not designed for incomplete and approximate queries/answers Hard to distribute Approximate answers to precise queries If the query is unsat, return the best almost sat solution found Precises answers to approximate queries Return a subset of existing solutions instead of showing them all griffioen SUM 2008 - October 2, 2008 8 / 24
  • 28. Problem and context Method proposed Experimental results Conclusion Motivation Designed to return results only when there are some Not designed for incomplete and approximate queries/answers Hard to distribute Approximate answers to precise queries If the query is unsat, return the best almost sat solution found Precises answers to approximate queries Return a subset of existing solutions instead of showing them all Interactive querying Use of intermediate results to help the user improving his griffioen query SUM 2008 - October 2, 2008 8 / 24
  • 29. Problem and context Method proposed Experimental results Conclusion 1 What’s the problem ? Querying RDF datastores Standard techniques 2 And Now for Something Completely Different Guessing the solution instead The way we do it 3 Does it work ? Evolution of the quality Some characteristics of this method 4 TODO list griffioen SUM 2008 - October 2, 2008 9 / 24
  • 30. Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : griffioen SUM 2008 - October 2, 2008 10 / 24
  • 31. Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables griffioen SUM 2008 - October 2, 2008 10 / 24
  • 32. Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book griffioen SUM 2008 - October 2, 2008 10 / 24
  • 33. Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid griffioen SUM 2008 - October 2, 2008 10 / 24
  • 34. Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid Triple Is in the graph ? <Ullman88> type Book yes <Ullman88> label Book no griffioen SUM 2008 - October 2, 2008 10 / 24
  • 35. Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid Triple Is in the graph ? <Ullman88> type Book yes <Ullman88> label Book no 3 If the solution is OK, stop. Otherwise, try again with something else griffioen SUM 2008 - October 2, 2008 10 / 24
  • 36. Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid Triple Is in the graph ? <Ullman88> type Book yes <Ullman88> label Book no 3 If the solution is OK, stop. Otherwise, try again with something else Rely on membership testing (instead of lookup) griffioen SUM 2008 - October 2, 2008 10 / 24
  • 37. Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid Triple Is in the graph ? <Ullman88> type Book yes <Ullman88> label Book no 3 If the solution is OK, stop. Otherwise, try again with something else Rely on membership testing (instead of lookup) The testing loop can be stopped at any time griffioen SUM 2008 - October 2, 2008 10 / 24
  • 38. Problem and context Method proposed Experimental results Conclusion Approach “I’m Feeling Lucky” approach : 1 Assign some random values to the variables ?publication = <Ullman88> ?title = Book 2 Verify if the solution is valid Triple Is in the graph ? <Ullman88> type Book yes <Ullman88> label Book no 3 If the solution is OK, stop. Otherwise, try again with something else Rely on membership testing (instead of lookup) The testing loop can be stopped at any time A result may satisfy part of the query griffioen SUM 2008 - October 2, 2008 10 / 24
  • 39. Problem and context Method proposed Experimental results Conclusion Our choices Need to pay attention to two aspects griffioen SUM 2008 - October 2, 2008 11 / 24
  • 40. Problem and context Method proposed Experimental results Conclusion Our choices Need to pay attention to two aspects 1 Each try should be a step closer to the solution Random guessing may never end Stopping the process at t + 1 should give better results than at t griffioen SUM 2008 - October 2, 2008 11 / 24
  • 41. Problem and context Method proposed Experimental results Conclusion Our choices Need to pay attention to two aspects 1 Each try should be a step closer to the solution Random guessing may never end Stopping the process at t + 1 should give better results than at t 2 Testing a candidate solution must be fast Will try a lot of solutions griffioen SUM 2008 - October 2, 2008 11 / 24
  • 42. Problem and context Method proposed Experimental results Conclusion Our choices Need to pay attention to two aspects 1 Each try should be a step closer to the solution Random guessing may never end Stopping the process at t + 1 should give better results than at t 2 Testing a candidate solution must be fast Will try a lot of solutions We made the following choices Generation of solutions : Evolutionary algorithm Verification of solutions : Bloom filter based testing griffioen SUM 2008 - October 2, 2008 11 / 24
  • 43. Problem and context Method proposed Experimental results Conclusion Binary Bloom filters (1/2) Compact representation of information : a set of n = 8 bits 1 2 3 4 5 6 7 8 Supports two operations INSERT ( KEY ) : Insert a key into the filter CONTAINS ( KEY ) : Test for the presence of a key Use k = 3 hash functions to compute a set of bits from a key HASH 1(“ HELLO WORLD ”)=8 HASH 2(“ HELLO WORLD ”)=6 HASH 3(“ HELLO WORLD ”)=3 griffioen SUM 2008 - October 2, 2008 12 / 24
  • 44. Problem and context Method proposed Experimental results Conclusion Binary Bloom filters (2/2) INSERT (“ HELLO WORLD ”) Current Bit-wise or operation OR Always successful (i.e. “Hello world” unlimited capacity) = Precision depends of New number of elements m. CONTAINS (“B ONJOUR !”) Current Bit-wise and operation AND “Bonjour !” Positive result can be a collision = kn griffioen Test result perror = (1 − e− m )k SUM 2008 - October 2, 2008 13 / 24
  • 45. Problem and context Method proposed Experimental results Conclusion A first (naive) approach Insert all the triples into a unique Bloom filter. INSERT (“<Ullman88>_type_Book”) INSERT (“<Ullman88>_label_"Principles of ..."”) ... griffioen SUM 2008 - October 2, 2008 14 / 24
  • 46. Problem and context Method proposed Experimental results Conclusion A first (naive) approach Insert all the triples into a unique Bloom filter. INSERT (“<Ullman88>_type_Book”) INSERT (“<Ullman88>_label_"Principles of ..."”) ... Use the CONTAINS operation to verify a solution CONTAINS (“<Ullman88>_type_Book”) ⇒ true CONTAINS (“<Ullman88>_label_Book”) ⇒ false griffioen SUM 2008 - October 2, 2008 14 / 24
  • 47. Problem and context Method proposed Experimental results Conclusion A first (naive) approach Insert all the triples into a unique Bloom filter. INSERT (“<Ullman88>_type_Book”) INSERT (“<Ullman88>_label_"Principles of ..."”) ... Use the CONTAINS operation to verify a solution CONTAINS (“<Ullman88>_type_Book”) ⇒ true CONTAINS (“<Ullman88>_label_Book”) ⇒ false Not the best approach ! Let’s see what happen in detail . . . griffioen SUM 2008 - October 2, 2008 14 / 24
  • 48. Problem and context Method proposed Experimental results Conclusion A first (naive) approach Insert all the triples into a unique Bloom filter. INSERT (“<Ullman88>_type_Book”) INSERT (“<Ullman88>_label_"Principles of ..."”) ... Use the CONTAINS operation to verify a solution CONTAINS (“<Ullman88>_type_Book”) ⇒ true CONTAINS (“<Ullman88>_label_Book”) ⇒ false Not the best approach ! Let’s see what happen in detail . . . ?publication label ?title CONTAINS (“<Ullman88>_label_Book”) griffioen modify ?publication and ?title SUM 2008 - October 2, 2008 14 / 24
  • 49. Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book griffioen SUM 2008 - October 2, 2008 15 / 24
  • 50. Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book SPO griffioen SUM 2008 - October 2, 2008 15 / 24
  • 51. Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book <Ullman88>_type SPO SP griffioen SUM 2008 - October 2, 2008 15 / 24
  • 52. Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book <Ullman88>_type type_Book SPO SP PO griffioen SUM 2008 - October 2, 2008 15 / 24
  • 53. Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book <Ullman88>_type type_Book <Ullman88>_Boo SPO SP PO SO griffioen SUM 2008 - October 2, 2008 15 / 24
  • 54. Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book <Ullman88>_type type_Book <Ullman88>_Boo SPO SP PO SO Three domains are defined S = <Ullman88> b1 ullman P = type label author _1 homepage O = Book "Principles of ..." b1 ullman <http://...> griffioen SUM 2008 - October 2, 2008 15 / 24
  • 55. Problem and context Method proposed Experimental results Conclusion Graph parsing Every triple of the graph is inserted into 4 Bloom filters <Ullman88> type Book <Ullman88>_type_Book <Ullman88>_type type_Book <Ullman88>_Boo SPO SP PO SO Three domains are defined S = <Ullman88> b1 ullman P = type label author _1 homepage O = Book "Principles of ..." b1 ullman <http://...> Each term is replaced by an integer (with a dictionary) griffioen <Ullman88> → 46 SUM 2008 - October 2, 2008 15 / 24
  • 56. Problem and context Method proposed Experimental results Conclusion Evolutionary algorithm flowchart [Eiben2003] Set of populations + Set of operators griffioen SUM 2008 - October 2, 2008 16 / 24
  • 57. Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title griffioen SUM 2008 - October 2, 2008 17 / 24
  • 58. Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title Creation of constraints to verify griffioen SUM 2008 - October 2, 2008 17 / 24
  • 59. Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title Creation of constraints to verify Clause ?publication type Book . bloom(spo |?publication1 type Book) bloom(sp |?publication1 type) bloom(po |type Book) griffioen SUM 2008 - October 2, 2008 17 / 24
  • 60. Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title Creation of constraints to verify Clause ?publication type Book . bloom(spo |?publication1 type Book) bloom(sp |?publication1 type) bloom(po |type Book) Clause ?publication label ?title . bloom(spo |?publication2 label ?title) bloom(sp |?publication2 label) bloom(po |label ?title) bloom(so |?publication2 ?title) griffioen SUM 2008 - October 2, 2008 17 / 24
  • 61. Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title Creation of constraints to verify Clause ?publication type Book . bloom(spo |?publication1 type Book) bloom(sp |?publication1 type) bloom(po |type Book) Clause ?publication label ?title . bloom(spo |?publication2 label ?title) bloom(sp |?publication2 label) bloom(po |label ?title) bloom(so |?publication2 ?title) Equality constraint equal(?publication1 ,?publication2 ) griffioen SUM 2008 - October 2, 2008 17 / 24
  • 62. Problem and context Method proposed Experimental results Conclusion Query parsing Definition of the chromosome for the individuals ?publication1 ?publication2 ?title Removed Creation of constraints to verify because Clause ?publication type Book . always true bloom(spo |?publication type Book) 1 bloom(sp |?publication1 type) bloom(po |type Book) Clause ?publication label ?title . bloom(spo |?publication2 label ?title) bloom(sp |?publication2 label) bloom(po |label ?title) bloom(so |?publication2 ?title) Equality constraint equal(?publication1 ,?publication2 ) griffioen SUM 2008 - October 2, 2008 17 / 24
  • 63. Problem and context Method proposed Experimental results Conclusion Evaluation of a candidate solution Solution is checked against all the constraints. If one is satisfied, A global reward w is won Each variable used is equally rewarded Rewards for : bloom(spo|?publication2 label ?title) reward(solution) += w w reward(?publication1 ) += 2 reward(?title) += w 2 griffioen SUM 2008 - October 2, 2008 18 / 24
  • 64. Problem and context Method proposed Experimental results Conclusion Creation of new individuals Select two individuals and do a one point crossover dblp:ullman <Ullman88> "Principles. . . " dblp:ullman <Ullman88> _:b1 <Ullman88> dblp:ullman _:b1 <Ullman88> dblp:ullman "Principles. . . " Randomly pick a pivot point Swap the two parts Mutate the least efficient variable dblp:ullman <Ullman88> "Principles. . . " 0 3×w 2×w <Ullman88> <Ullman88> "Principles. . . " Select the variable with lowest Assign a random new value reward griffioen SUM 2008 - October 2, 2008 19 / 24
  • 65. Problem and context Method proposed Experimental results Conclusion 1 What’s the problem ? Querying RDF datastores Standard techniques 2 And Now for Something Completely Different Guessing the solution instead The way we do it 3 Does it work ? Evolution of the quality Some characteristics of this method 4 TODO list griffioen SUM 2008 - October 2, 2008 20 / 24
  • 66. Problem and context Method proposed Experimental results Conclusion Results on some (small) datasets Database FOAF (15k triples) and DBLP (3M triples) Query with, respectively, 4 and 11 different variables Average result for 200 individuals and 500 generations 60 100 50 90 fitness value fitness value 40 30 80 20 70 10 0 60 0 100 200 300 400 500 0 100 200 300 400 500 n-th generation n-th generation Solutions with maximum reward (52) are found for FOAF griffioen Not enough time for DBLP (max 319) SUM 2008 - October 2, 2008 21 / 24
  • 67. Problem and context Method proposed Experimental results Conclusion Scalibility & speed Low memory requirements Only depends on the number of individuals and the size of the Bloom filters (a) parsing (b) querying dataset memory dataset memory FOAF 65 MB FOAF 15 MB DBLP 230 MB DBLP 140 MB Table: Average memory usage (mostly due to dictionary) Computation can be distributed Candidate solutions are independent The dictionary can be based on a DHT griffioen SUM 2008 - October 2, 2008 22 / 24
  • 68. Problem and context Method proposed Experimental results Conclusion 1 What’s the problem ? Querying RDF datastores Standard techniques 2 And Now for Something Completely Different Guessing the solution instead The way we do it 3 Does it work ? Evolution of the quality Some characteristics of this method 4 TODO list griffioen SUM 2008 - October 2, 2008 23 / 24
  • 69. Problem and context Method proposed Experimental results Conclusion Status and future work Current status The search process can be slow to converge Several parameters to tune (rewards, size of the population, number of generations, . . . ) griffioen SUM 2008 - October 2, 2008 24 / 24
  • 70. Problem and context Method proposed Experimental results Conclusion Status and future work Current status The search process can be slow to converge Several parameters to tune (rewards, size of the population, number of generations, . . . ) Current work griffioen SUM 2008 - October 2, 2008 24 / 24
  • 71. Problem and context Method proposed Experimental results Conclusion Status and future work Current status The search process can be slow to converge Several parameters to tune (rewards, size of the population, number of generations, . . . ) Current work 1 Improve benchmarking Test with more queries and more datasets Better study of the influence of the parameters griffioen SUM 2008 - October 2, 2008 24 / 24
  • 72. Problem and context Method proposed Experimental results Conclusion Status and future work Current status The search process can be slow to converge Several parameters to tune (rewards, size of the population, number of generations, . . . ) Current work 1 Improve benchmarking Test with more queries and more datasets Better study of the influence of the parameters 2 Improve evolution Experiment different type of crossover and mutation Implement dynamic valuations for the rewards Improve early results on tabbu search approach griffioen SUM 2008 - October 2, 2008 24 / 24
  • 73. Problem and context Method proposed Experimental results Conclusion Status and future work Current status The search process can be slow to converge Several parameters to tune (rewards, size of the population, number of generations, . . . ) Current work 1 Improve benchmarking Test with more queries and more datasets Better study of the influence of the parameters 2 Improve evolution Experiment different type of crossover and mutation Implement dynamic valuations for the rewards Improve early results on tabbu search approach 3 Test other, easy to parallelize and anytime, optimizer Swarm based algorithm (PSO, ...) or an other EA griffioen CSP solver SUM 2008 - October 2, 2008 24 / 24