:class:`kgextension.link_exploration.link_explorer`
The Link Exploration allows to extensively search the Linked Open Data (LOD) Cloud for URIs connected to the original URIs in the column obtained from linking (see :ref:`linker-label`). It follows the defined links starting from a base link to a certain number of hops. By default it follows the owl:sameAs predicate to hop from link to link (it recursively applies the predicate on the newly obtained links). It adds the discovered links as new columns to the dataframe.
Consider as an example input the following toy example: For a dataset entity "Berlin", the linker identified the corresponding DBpedia URI "http://dbpedia.org/resource/Berlin".
| city | uri |
|---|---|
| Berlin | http://dbpedia.org/resource/Berlin |
The link explorer can then be used to identify URIs belonging to Berlin as well and add them to the DataFrame:
The linker traverses through the different avaiable knowledge graphs as well as through the different language versions of DBpedia. Note that this is a shortened example: with a large number of hops specified, the link explorer can find many more URIs.
In its default version, the link explorer performs one hop following the owl:sameAs predicate without filtering.
df_link_explored = link_explorer(
df, base_link_column, number_of_hops = 1,
links_to_follow = ["owl:sameAs"],
lod_sources = [], exclude_sources = []
)In order to generate more URI columns, the number of hops can be increased to any positive integer. This means that if we set the number of hops to two, we follow e.g. the link of the first hop to the French wikipedia URI of Berlin, http://fr.dbpedia.org/resource/Berlin, and then follow all other owl:sameAs links from this French site again. If the number of hops were three, and we followed among all links of the French site also the one to the Greek site, we would again hop on all other links occurring on the Greek dbpedia URI of Berlin.
df_link_explored = link_explorer(
df, base_link_column='uri', number_of_hops = 2
)Warning
Increasing the number of hops by one integer point, e.g. from 2 to 3, does not imply that three columns with URIs will be added, but that from the first resource all links will be followed, which in turn will all be the starting points for the next hop and so on. Thus, even a small number of hops is approximately exponential in time.
The link explorer can follow arbitrary links and also multiple at the same time. In this example, we add rdfs:seeAlso to the list of links to follow:
df_link_explored = link_explorer(
df, base_link_column='uri',
links_to_follow=["owl:sameAs","rdfs:seeAlso"]
)For the http://dbpedia.org/resource/Berlin, now URIs that are connected to it via the rdfs:seeAlso property are added as well, e.g. http://dbpedia.org/resource/Sister_city.
Since the process of hopping from link to link can be quite randomly, it is possible to specify sources the results should be limited two, e.g. certain knowledge graphs. Use strings or regular expressions to define, which sources to include. In the following example, the links are limited to the nytimes and geonames knowledge graphs.
df_link_explored = link_explorer(
df, base_link_column='uri',
lod_sources=["nytimes","geonames"]
)In contrast to specifying certain sources, it is also possible to define knowledge graphs to be absent from the newly generated URIs. Use strings or regular expressions to define, which sources to exclude. In the following example, URIs from dbpedia will not be allowed.
df_link_explored = link_explorer(
df, base_link_column='uri',
exclude_sources=["dbpedia"]
)