1 Introduction
2 Related work
2.1 Natural language querying
2.2 Graph-based querying
2.3 Qualitative comparison of the selected semantic search systems
System | User query interface | Domain | Allow keyword search | Type and property recommendations |
---|---|---|---|---|
Facebook Graph Search
| Natural language | Facebook | Yes | No |
NAGA
| Graph-based | YAGO | Yes/No | No |
SFC
|
Graph-based
|
Any
|
Yes
|
Yes
|
GoRelations
| Graph-based | DBpedia | No | No |
PANTO
| Natural language | Any | Yes | No |
PowerAqua
| Natural language | Any | Yes | No |
NITELIGHT
| Graph-based | Any | Yes | No |
SPARK
| Keyword-based | Any | Yes | Yes |
System | Aggregation | SPARQL query types | Order by | Pagination | Optional | Filter | Union |
---|---|---|---|---|---|---|---|
NAGA
| No | Select only | No | Yes | No | No | No |
SFC
|
No
|
Select only
|
No
|
Yes
|
Yes
|
Yes
|
No
|
GoR
| No | Select only | No | No | No | Yes | No |
PANTO
| No | Select only | No | No | No | Yes | Yes |
NITELIGHT
| Yes | Select only | Yes | Yes | Yes | Yes | Yes |
3 Concept of Semantic Focused Crawler
-
recommendation of variables automatically associated with appropriate types allowing to create more precise queries in a simpler manner,
-
variable properties enabling to discover relations connecting given objects,
-
recommendation of types and relations allowing to avoid error-prone mapping between query terms and ontology types or properties.
3.1 Query construction process
-
Claudia Schiffer has in height \(x_1\).
-
\(x_2\) is a Model.
-
\(x_2\) has in height \(x_1\).
3.2 Recommendation of variable names
-
\(x_1\) is a FootballPlayer. \(x_2\) is a Actress.
-
\(x_1\) has wife \(x_2\) . \(x_1\) plays in \(x_3\).
-
\(x_3\) is a FootballClub.
3.3 Variable properties
3.4 Recommendation of object types and relations
4 Evaluation
4.1 Evaluation methodology assumptions
4.2 QALD query set
4.3 Accuracy measurement
4.4 Query complexity measurement
System | Semantic representation of the query | Number of elements |
---|---|---|
GoR
|
Julia Roberts/Actor, starredIn, \(x_{1}\)/Movie
| 9 |
Garry Marshall/Director, directed, \({x_1}\)
| ||
SFC
|
Garry Marshall, directed, \(x_1\)
|
5
|
Julia Roberts, starredIn, \({x_1}\)
| ||
NAGA
|
Julia Roberts, actedIn, \({x_1}\)
| 6 |
Garry Marshall, directed, \({x_1}\)
|
5 Evaluation results
System | Precision | Recall |
F-score |
---|---|---|---|
SFC
|
0.961
|
0.906
|
0.927
|
GoR
| 0.693 | 0.672 | 0.682 |
NAGA
| 0.661 | 0.639 | 0.649 |
Google (Natural Language)
| 0.620 | 0.632 | 0.626 |
PowerAqua (DBpedia only)
| 0.512 | 0.481 | 0.494 |
Google (Keywords)
| 0.422 | 0.425 | 0.423 |
System | SFC | GoR | NAGA |
---|---|---|---|
Elements per query
|
2.786
| 5.714 | 3.571 |
6 Conclusions
6.1 Main results
-
Keyword-based search as a function of the graph-based search interface.
-
Intuitive variable naming convention which allows the user to creates triples which are similar to simple sentences.
-
Automatic type and property suggestions that the user may use in the query construction process.