=Paper=
{{Paper
|id=Vol-2180/paper-61
|storemode=property
|title=Towards a Semantic Discovery for Heterogenous Open Data by Interlinking Metadata of
Datasets
|pdfUrl=https://ceur-ws.org/Vol-2180/paper-61.pdf
|volume=Vol-2180
|authors=Jiseong Son,Youngsung Son,Haklae Kim
|dblpUrl=https://dblp.org/rec/conf/semweb/SonSK18
}}
==Towards a Semantic Discovery for Heterogenous Open Data by Interlinking Metadata of
Datasets==
https://ceur-ws.org/Vol-2180/paper-61.pdf
Towards a Semantic Discovery for Heterogenous
Open Data by Interlinking Metadata of Datasets
Jiseong Son1 , Youngsung Son2 , and Haklae Kim1
1
Korea Institute of Science and Technology Information, Korea
{jsson,haklaekim}@kisti.re.kr
http://www.kisti.re.kr
2
Electronics and Telecommunications Research Institute
{ysson}@etri.re.kr
Abstract. Open data refers to data that everyone can freely use, reuse
and redistribute. A number of open data is released by various organi-
sations, governments or communities. However, it is limited to discover
datasets that users want, since most of data portals allow to search their
datasets based on simple keywords using file names or descriptions, etc.
This paper proposes a novel way for discovering disclosed government
datasets by using linked data technologies. For achieving this objective,
a set of datasets is collected from the public data portal in Korea, and
all of data fields are extracted and transformed into linked data using an
ontology model. We also provide a simple evaluation, which compares a
search performance between the portal and the proposed method.
Keywords: Open Data · Government Data · Semantic Discovery · On-
tology.
1 Introduction
While the big data phenomenon is becoming increasingly common, it is not easy
for anyone to freely use the data. A large amount of big data is owned by service
providers or platform owners, and only a limited portion of data is shared. On the
other hand, open data allows users to provide a significant opportunity that they
are able to use a variety of data across heterogeneous data sources and domains.
The key value of open data is that a piece of data contained in published data
can be interlinked with other data. In an open data environment, data can
be interchanged between institutions, between institutions and governments, or
between governments, and new value can be created through interlinking of
datasets [2].
One of issues aligning on open data is that discovering datasets is getting
difficult [1]. Most data portals provide the ability to discover datasets. For ex-
ample, CKAN (Comprehensive Knowledge Archive Network), which is a data
portal platform, is able to retrieve a file name and a description for files, and
tags and file types added to the dataset [5, 6]. However, there is a limitation
to searching for the information that an individual dataset has. If a user wants
Fig. 1. A data model of representing a dataset and its metadata
to find out datasets that have ‘population’, most of data portals returns a list
of datasets that contains the keyword (i.e. ‘population’) on descriptions or file
names of the datasets on behalf of retrieving their content [4].
This study proposes a method of discovering disclosed government datasets
by extracting data fields of individual datasets and constructing them as linked
data. Section 2 describes a research approach including data collections and
transformations based on a proposed ontology model. Section 3 introduces a
small evaluation to retrieve the collected datasets with some comparisons. Sec-
tion 4 concludes and introduces future research.
2 Research approach
We collect a set of public open datasets from the public data portal3 and extract
all of data fields from the datasets. This site provides governmental open datasets
released by the Republic of Korea. Currently, 689 organisations provide 22, 334
file data (CSV or other types), 2,547 open APIs, and 91 standard data.
This paper focuses on analysing the standard data, since metadata quality of
other datasets is not good to our purposes [3]. Note that the standard data in the
portal refers to a set of datasets by using the public data open standard guidelines
of the government that defines an item name (data field) and its value for 93
domains. A total of 1,480 item names were extracted in the collected standard
datasets, there are 903 item names that eliminate redundancy. The selected fields
are no needs for further clustering, since a data field is already normalised by
using standardised terms. Note that the collected datasets containing the road-
name address and the land-number address are 53 and 44, respectively, and the
latitude and longitude include 55 data sets. Latitude and longitude data fields
are defined together in all datasets. There are 5 cases where latitude/longitude
is included in the dataset in which the road-name address as an item name does
3
http://data.go.kr
Fig. 2. The search results for the data portal and the proposed method
not exist, and there are 12 cases when there is no land-number address. On the
other hand, when there is no latitude and longitude item name, the road-name
address and the land-number address correspond to one case of three. There are
14 datasets that do not have both address and latitude/longitude information.
A simple ontology model is designed for representing a relationship between
a dataset and its data field as shown in Figure 1. Each dataset has a set of data
fields, and this relation is represented by using the data:hasDataItem property.
Note that the data:relatedTo property is to describe a relationship between spe-
cific terms. For example, a ‘location’ may be related to ‘address’, ‘latitude’, or
‘longitude’. There is no dataset with an item name of ‘location’, but most of
datasets have ‘address’ or ‘latitude and longitude’. In this reason, this property
is used to expand a specific query. As shown in Figure 1, a traditional market
dataset does not have any fields associated to a toilet. However, it is possible to
discover some toilets around a traditional market, because both datasets have
address or locational information.
3 Evaluation
We report the measurements obtained in Figure 2. We compare the three cases
for the data portal and the proposed method. Case 1 discovers for a dataset
with a single keyword. The portal and the proposed model have 1 and 3 results,
respectively, for a specific topic (i.e. ‘toilet’). Two of the results of the proposed
model have no related keywords in the file name or description. Case 2 is a
method for searching heterogeneous datasets. Consider the following query: what
datasets contain a market and toilet information nationwide? Such queries are
dependent on the information contained in the dataset. Although a particular
dataset can be discovered if it has both fields, searching in a fragmented dataset
is difficult. As shown in Figure 2, the portal does not have search results for
multiple keywords (i.e. ‘market’ and ‘toilet’), but the proposed model gives two
results. However, these results provide a simple information about a toilet as yes
or no. Case 3 is to find out a specific relationship between datasets. For example,
the data:relatedTo property can be used for discovering a relationship between a
traditional market and a toilet. First, it retrieves a list of exact administrative
area from both datasets based on address information, and then calculates a
distance between search results using the latitude and longitude information.
Compared to Case 2, this result show a specific location of a toilet around the
market.
4 Conclusion
This paper proposes a new approach to discover datasets on a data portal by
using linked data technologies. Most of data portals allow users to retrieve their
datasets with search options, including keywords, data types, or user-generated
tags, etc. However, it is limited to discover datasets based on their content. In
this reason, users need to check whether these datasets are suitable to their pur-
poses about search results. To solve this problem, this paper introduces a simple
semantic search that aims to discover internal content of individual datasets by
constructing linked data including data fields from individual datasets and its
relationships. Although experimental data are relatively small, the evaluation
shows that the proposed method is more effective than existing search methods.
Future research will apply the data model and search method proposed in this
paper to the whole data provided by the public data portal.
References
1. Hand, D.: Data, not dogma: Big data, open data, and the opportunities ahead. In:
Tucker, A., Hppner, F., Siebes, A., Swift, S. (eds.) Advances in Intelligent Data
Analysis XII, Lecture Notes in Computer Science, vol. 8207, pp. 1–12. Springer
Berlin Heidelberg (2013)
2. Janssen, M., Charalabidis, Y., Zuiderwijk, A.: Benefits, adoption barriers and myths
of open data and open government. IS Management 29(4), 258–268 (2012)
3. Kim, H.: Quality evaluation of the open government data: The case of the open
data portal of korea. International Journal of Contents (in press)
4. Kostovski, M., Jovanovik, M., Trajanov, D.: Open data portal based on semantic
web technologies. In: Proceedings of the 7th Annual South-East European Doctoral
Student Conference (DSC 2012). pp. 504–516 (2012)
5. Loureno, R.P.: Evidence of an open government data portal impact on the public
sphere. IJEGR 12(3), 21–36 (2016)
6. Scholz, R., Tcholtchev, N., Lmmel, P., Schieferdecker, I.: A ckan plugin for data
harvesting to the hadoop distributed file system. In: Ferguson, D., Muoz, V.M.,
Cardoso, J.S., Helfert, M., Pahl, C. (eds.) CLOSER. pp. 19–28. SciTePress (2017)