In this paper, we present a conceptual study aimed at understanding the impact of international resource synchronization in Wikipedia and DBpedia. In the absence of any information synchronization, each country would construct its own datasets and manage it from its users. Moreover the cooperation across the various countries is adversely a ected. The solution is based on the analysis of Wikipedia infobox templates and on experimentation such as term translation.
Wikipedia is an international project which is a web-based, free-content encyclopedia and is written collaboratively. Wikipedia has made tremendous e ect in the web. It has grown rapidly into one of the largest reference web sites. The main advantage of using Wikipedia is its wide coverage of concepts and languages. Wikipedia currently comprises more than 260 languages. However, Wikipedia still lacks sufcient support for non-English languages. The 22% articles in all Wikipedias belong to the English language version. Although English has been accepted as a global standard to exchange information between di erent countries, companies and people, the majority of users are attracted by projects and web sites if the information is available in their native language as well. One can also assume that the proportion of users on the Internet who do not speak English will continue to rise.
Due to the di erences in the amount of information between English and non-English languages in Wikipedia, there needs to not only avoid information loss, but also enrich informations.
Copyright is held by the author/owner(s).
WWW2010, April 26-30, 2010, Raleigh, North Carolina.
In this paper we presents our e orts to create a
multilingual system for translation-based information
synchronization. Our work is based on idea that Wikipedia as a
multilingual corpus and especially translation results
between di erent editions provide valuable multilingual
resources to the web. To explore this problem, we developed
Metadata Synchronization, a platform for translation-based
data synchronization between English Wikipedia and
Korean Wikipedia. It aims to translate infoboxes from the
English Wikipedia into Korean and insert it into the Korean
Wikipedia. Because Wikipedia o ers a number of structural
elements, in particular, the infobox template is used to
express structured information about a condensed set of
important facts relating to the article[
The rest of this paper is organized as follows: In section 2, we describe related work. The framework and details of the proposed approach are given in section 3. Section 4 and 5 discusse the experimentation and results. At the end, we summarize the obtained results and point our future work. 2.
Wikipedia represents a valuable source of knowledge to
extract semantic information between concepts. [
In [
DBpedia1[
[
In the area of cross-language data fusion, another project
has been launched[
Our metadata synchronization algorithm consists of two consecutive steps: a metadata extraction step and a metadata translation step. Figure 1 illustrates the work ow of our approach. In order to illustrate our approach, we examine a case study in Wikipedia by using the Infobox as metadata. Wikipedia is considered to be one of the most successful collaborative editing communities on the web, we consider it as an interesting example to discuss multilingual synchronization.
In particular, Web pages pose main problems to multilingual resource: The S-group contains pairs of articles which use the same infobox template but have a di erent amount of information. For example, an English-written article and a non-English-written article, which have an interlanguage link and use the same infobox template, but have a di erent amount of template attributes. The D-group contains pairs of articles which use different infobox templates. D-group emerges due to the di erent degrees of each Wikipedia communities' activities. In communities where many editors lively participate, template categories and formats are more wellde ned and more ne-grained. For example, philosopher, politician, o ceholder and military person templates in English are matched just person template in Korean. It appears not only a Korean Wikipedia but also non-English Wikipedias.
The M-group contains pairs of articles where an infobox exists on only one side.
NRHP and MLB player. In the cases of SW books and NRHP, these stand for Star Wars books and National Register of Historic Places respectively. However, a SW is widely used for an abbreviation of a software and NRHP is used only in Wikipedia. Thus it is di cult to understand.
The second reason is that the template name does not
have same meaning as the common sense in a dictionary.
For example, according to the Wiktionary[
Infobox television=fshow name, image, caption, genre, format, creator, developer, writer, director, producer, country, opentheme, ...g
Moreover di erent format of properties with same meaning should be re ned. In order to overcome the problems of multiple property names being used for the same type, for example, `birth place', `birthplace and age' and `place birth' are mapped to `birthPlace'. In addition, the `img' property is replaced into the `image', because the later form is better to translate. If this is solved, it will be a great help for organizing templates and the e ciency of our synchronization framework will be higher.
We extracted English triples in infoboxes are simply
translated into Korean triples. For the dictionary-based term
translation [
After the translation step, we try to align the bilingual metadata. For the alignment, we manually constructed the mapping table between Korean and English template schemas based on translation results.
The details of this method are described in the following. 4.
We introduce the details of experimental dataset and the processing of the data. 4.1
We need two types of data sets for experimentation. One type is the Wikipedia dump and the other type of data is the DBpedia dataset. In this paper, we only use Korean and English data, but our approach can be applied on any language of data in Wikipedia. We extracted 37,576 Korean articles contain infobox using Wikipedia dumps at March, 2010. Once the data is ready, all articles are parsed to extract attribute-value pairs of infobox. We got 1,042 infobox templates in Korean. We got 2,792 infobox templates in English using DBpedia dataset, and 1,642 template-pairs. It is noted lots of templates are duplicate uses. Thus, the number of template-pairs is much bigger than the number of templates in Korean. However, there are many articles having infobox without the name of template, it is a hurdle in the progress and development of the infobox extraction. 4.2
We executed the translation from English triples in infoboxes to Korean triples. In our experiments, we used DBpedia 3.4 as resource for the translation, a comparison of datasets is as follows:
We can get translated Korean triples over 30 times larger than existing Korean triples. However, its quality is still quite poor. Because lots of triples include sentences or phrases such as below:
Triple 1. \%21%21%21", \currentMembers", \Nic Offer Allan Wilson Mario Andreoni Tyler Pope Dan Gorman Sean McGahan Shannon Funchess Paul Quattrone" Also, a large amount of translated triples consists of only numbers. Our translation approach is compared to a statistical machine translation system by Google Translate API. The Google Translate API is an initial prototype used a statistical MT system based on Moses and trained on Europarl. Overall, the Google Translate API performed better and executed faster than our dictionary based translation. However, in case of proper nouns and abbreviations, our approach yields slightly higher accuracy than Google Translate API. 5.
Today, we created the metadata using translated results. However, we did not check the consistency between existing things and new things. To solve this problem, we try to utilize this consistency management to construct a large and ne-grained ontology by using infobox template. Thus we have built a template ontology, OntoCloud4, from DBpedia and Wikipedia to e ciently build the template structure.
The construction of OntoCloud consists of the following steps: (1) extracting templates of DBpedia as concepts in an ontology, for example, the Template:Infobox Person (2) extracting attributes of these templates, for example, name of Person. These attributes are mapped to properties in ontology. (3) constructing the concept hierarchy by set inclusion of attributes, for example, Book is a subclass of Book series. For the ontology building, similar types of templates are mapped to one concept. For example, the Template:infobox baseball player and Template:infobox asian baseball player describe baseball player. Using the OntoCloud, we could be check the consistency of templates.
As the web grows in number of pages and amount of information, there is an increasing interest towards supporting tasks such as organizing, and enriching. We have proposed a novel idea on using term translation not only to synchronize but also to enrich information of multilingual web resources, and presented an e ective approach to implement this idea in Wikipedia. Our work is ongoing for technical improvements, such as better alignment between bilingual metadata, and more precise translating. After the veri cation of template consistency, the Korean Wikipedia and DBpedia can be updated automatically. This will be helpful to guarantee that the same information can be recognized in di erent languages. Moreover, will be helpful to edit articles and to create infoboxes when a new article is created. It can support the authors by suggesting the right template. As future work, it is planned to support more standardization for the Korean language and improve the quality of the translated datasets.