Wikipedia is one of the most valuable language and ontological resources covering wide domains so that DBpedia, LOD based on Wikipedia, plays the important role in the LOD cloud by connecting various resources. DBpedia Japanese is the LOD created from Wikipedia Japanese just as DBpedia data sources in other languages like English and German. We here describe how the conversion could be carried out with the efforts to t DBpedia software and show the results of the dataset. We also describe how the created DBpedia Japanese is used by other Linked Data and show the Linked Data Cloud in Japanese.
2
DBpedia Japanese is the DBpedia generated from the Japanese Wikipedia. It is an internationalization of DBpedia where all parts fo the software used to build it from Wikipedia are those developed for English DBpedia. Therefore most of the building process were done except some language-speci c treatments like the con guration for information extraction, and ontology mapping.
DBpedia Japanese is built and maintained in the activity of LODAC Project5 where various data resources such as museums and biology are published as Linked Data.
Currently it contains 79,423,068 triples including links to the Japanese
WordNet and the Japanese Wikipedia Ontology[
DBpedia Information Extraction Framework (DIEF) 6 is the package of the
software to extract information from Wikipedia. It is basically applicable to
Wikipedia sources in any languages but is needed to customize in order to adapt
information written in the language[
)" HomepageExtractor This module is used to extract official web pages. We added three patterns to process for hyperlinks with the following strings. propertyNamesMap "ja" -> Set("homepage", "website", "
", " ") externalLinkSectionsMap
", "
", "Web "ja" -> " officialMap "ja" -> " "
" They mean Web page, Outgoing page from Wikipedia, and Official respectively. ImageExtractor This module is used to extract links to images. To avoid non re-usable images, we added the following line;
"ja" -> """(?i)\{\{\s?(Non free|Non-free pubart)\s?\}\}""".r PersondataExtractor This module is used to extract information on persons. We added \ja" in the supported languages and de ned the following patterns. 1. Names of templates for personal information 2. \ "(name) 3. \ "(alias) 4. \ "(abstract) 5. dates and places for birth and death
DBpedia Japanese uses IRIs to identify resources now so that there are no prolems to include Japanese characters. But there was the problem for coding since URI instead of IRI was used in Virtuoso before version 6.1.4. 2.2
The results of the conversion is shown in Table 1. It is the statistics by the rst build of DBpedia Japanese on January, 2012. Wikipedia Japanese contains 1,558,754 fragments including article, templates, image descriptions, and primary meta-pages. We can conclude ca. 90% of the original data is converted into DBpedia. The bold gures indicate the improvement by the con guration for Japanese Wikipedia. 2.3
In DBpedia, an ontology is created and shared among different languages. A class in the ontology typically corresponds to templates which are used to generate Infobox in Wikipedia. Properties in a class corresponds to parameters in templates. Ontology mapping in DBpedia is to create classes and their properties and to create mapping from templates in Wikipedia. Ontology mapping is not easy task indeed since it is required to understand both background knowledge in domains and structures in templates. In order to stimulate the ontology mapping activity, we hold two so-called \mapping party" where people gathered and created mapping together on August 2012 and March 2013 in which 10 and 25 people participated. The current results of mapping is shown in Table 2. The numbers of mapped templates are still low in comparison with English DBpedia. One of the reasons is that we did not create new classes yet, rather added mapping from the existing classes to templates in Wikipedia Japanese. There are some original classes like \ " or Samurai but we have not created such classes yet. We need more improvement on ontology mapping.
Japanese rate of all templates in 4.67% (81 Wikipedia are mapped of 1,733) rate of all properties 2.47% (1581 Wikipedia are mapped of 62,679) rate of all template occurrences 47.99% (286,858 Wikipedia are mapped of 597,696) rate of all property occurrences 38.75% (3,128,208 Wikipedia are mapped of 8,071,982) In this section, we describe Linked Data resources in Japanese. Here Linked Data resources in Japanese means Linked Data resources where their important or signi cant part is data presented in Japanese or data is mainly about Japan. An example for the former is DBpedia Japanese and the latter is the statistics data about Japanese Industry. The distinction between them is sometimes important but they mostly overlapped in our case since Japanese is mostly used in Japan.
Linked Data resources in Japanese have been limited but have increased rapidly in a couple of years. As mentioned in Section 1, LODAC project initiated to create datasets Linked Data in Japanese. We create and maintain LODAC Location, LODAC Museum, LODAC SPECIES, and WorldNet-J, and connect them to each other.
Semantic Web Community in Japan started the contest to promote LOD, namely the LOD Challenge. LOD Challenge was hold twice in 2012 and 2013, and collected over 70 and 200 applications respectively. It should be noted that the category of LOD challenge includes Dataset section as well as Idea and Application sections. Through the LOD challenge, several Linked Data resources are published.
We collect major Linked Data resources in Japan and map them in a gure (see Figure 3 and 1). The criteria to include datasets in it is as follows; 1. providing more than 10,000 triples, 2. providing either derefernece, data dump or SPARQL Endpoint, 3. providing labels in Japanese, and 4. open license is recommended but not mandatory.
The last criteria may seem odd but we think currently that forcing clear open licenses is not suitable at the beginning of the emergent community of LOD in Japan since there are a lot of datasets which are widely used just like with open licenses but do not have licenses yet. 3.1
The Linked Data Cloud in Japanese is still small, i.e., only 21 datasets are included which is just one tenth of Global LOD cloud. The proportion of the cloud is similar. Major category is publication-related datasets. Then some datasets are from geographic, life science and government domains. The clear difference is that there are many datasets without open licenses. It is mainly because the original datasets from which Linked Data datasets are generated often lack licenses. Some look very open but there are no clear mentions for licenses yet.
We pick up some of the datasets that are important resources connected to DBpedia tightly. 3.2
The efforts for multilingual WordNet has been made worldwide based on the
Princeton's English WorldNet so far. In 2008, the Japanese WordNet
(WNja) was developed and released by the National Institute of Information and
Communications Technology (NICT) in Japan [
There are some attempts to convert WorldNet into RDFs. In 2006, W3C
published the Working Draft for the representation in RDF of WorldNet 2.0 [
Since both WorldNet and Wikipedia are the most famous comprehensive
languages resources, there are many studies how combining them can contribute
to build better languages resources (e.g., Yago [
Table 3 shows the statistics of linking data of WN-ja to DBpedia Japanese, and Table 4 shows the statistics of linking data of DBpedia Japanese to WNja. In this attempt of linking by words in WN-ja and resource and property names in DBpedia Japanese, we made the connection by literally exact matching. Therefore, the mapping is exactly one by one and inversely equivalent in this case. DBpedia # of linked # of WN nouns rate resources 33,017 65,788 50.1% properties 1,245 65,788 1.9% DBpedia # of linked # of IRIs rate resources 33,017 1,456,158 2.3% properties 1,245 16,020 7.8% 3.3
Takahira Yamaguchi and his group.[
We have worked to create DBpedia Japanese as LOD resources in Japanese which can be used by other LOD datasets. What we have learnt through the process are as follows; 1. The language-dependent software problems are been solved mostly, but some still exist. They are not serious but it takes time and techniques to solve them. In particular, some software and tools are not completely compatible with IRI or UTF-8, and some lack language tags when describing literals. Anyway it is important to share demands from local communities with the global community. 2. People are willing to link to our resources. Within a year and so, we have 13 datasets which link their resources to those in DBpedia Japanese. The problem is that it is difficult to nd datasets linking to us. We collect the datasets by personal communication or by local academic meetings. We need some techniques to nd them automatically. 3. Having the hub for linking is not just good to make LOD cloud but also good to connect people. We had several academic meetings and tutorials for LOD. DBpedia Japanese is always a good example to demonstrate LOD since it covers various topics including daily issues and they are of course written in Japanese. People were motivated to join the LOD community in order to connect their data to DBpedia Japanese. 5
Wikipedia isthe precious ontology resources covering wide domains so that DBpedia play an important role in the LOD cloud by connecting various LOD resources. Building DBpedia for an lanugage is not so difficult task since all parts of the software are ready to use except some minor issues. We strongly recommend people in other languages to build own DBpedia. Building DBpedia is not just adding an useful resource but stimulating people to connect their data. It is the rst step to make the society to be connected by data. Open DATA METI Statistics KAKEN CiNii Biomass pedia Freebase
LSJ
PinQA
GeoLOD
VIAF DBpedia
DBpedia
Japanese LODAC Species Japanese Wikipedia Ontology
WordNet-ja
Yokohama Art LOD save MLAK
NDL Authority
Earthquake Archives Fukushima
LC Geonames
SOCIA Industry Life Science Media Publication User generated content
Geographic Cross-domain Government
LOD Cloud Open license
Fumihiro Kato, 2013-10-09 Kyoto Manga
Museum Aozora Bunko
RIHN Neji LOD LODAC
Museum