In this study, we analyzed and discussed the network of structured knowledge for data utilization using data jackets (DJ). A DJ is a method used to describe the summary information related to data, and its purposes are to improve the readability of the data by the users and to promote data understanding. The knowledge for data utilization consists of three-partite graphs that correspond to the three tuples, including the requirements, solutions (proposals of data usage), and DJs. By combining these units of knowledge, we can describe complex knowledge based on data utilization. Application of the methods of network analysis to the knowledge base allows us to understand the latent connections between data and knowledge elements that cannot be understood in the conventional network of data.
In view of the worldwide trend on big data and AI,
crossdisciplinary data exchange and collaboration constitutes one
of the essential social demands. Although platforms used to
display data information and their exchange have been
developed, a knowledge base intended to be used for a) the
promotion and the discovery of valuable data and b) for the
use of exchange methods for data utilization, has not been
well developed yet. To create a collaborative knowledge
base for data in this study, we have created a network of
structured knowledge for data utilization and have discussed
the results based on the analyses of its structure and features.
In this study, we used the results generated in the workshops
of Innovators Marketplace on Data Jackets (IMDJ) as the
knowledge elements for data utilization. The IMDJ was a
workshop conducted to analyze and discuss
cross-disciplinary data utilization using Data Jackets (DJs). A DJ is a
method used to summarize information regarding data, and
its purpose is to improve the readability of the data by the
users, and to promote data understanding
The network consisted of a three-partite graph. Figure 1 shows the largest component of the network. Hayashi and Ohsawa (2018b) used the terms in the data outlines in each TrafficvolumedaLtiafelogdWateaatherWdaetDabistPaesxutterrcdhaIatnasfeorhmisatotrioynofSroectiaaillemredSimaRadaraittlpawhaoynceRodnoagateadstairoenadinatYaokohama Figure 2: The top 10 DJs used for creating Solutions DJ as the contexts to create the context-based data network. Although the outlines included the intentions of the data acquisition and backgrounds of data usage, the context defined by their research was the context within the existing data. For cross-disciplinary data exchange and collaboration, it is necessary to consider the newly given contexts to combine data from different areas. Therefore, in this study, we defined the solutions and requirements that corresponded to the upper layer as the contexts, and not the terms in the DJ. Figure 2 shows the top 10 DJs used to creating solutions equivalent to the degrees of DJs in the network. Traffic volume data is most frequently used. Accordingly, three out of the top 10 DJs are related to traffic. The result suggests that data related to traffic are the hubs in the network. Furthermore, data derived from personal information—life log data, social media, and smartphones—are also frequently used (three out of the top 10 DJs).
Table 1 lists the basic features of the largest component
of the network. Given that it is a three-partite graph and
there is no triad in the network, the average degree is
relatively low, the density is extremely low, and the clustering
coefficient is zero, compared to a random network.
Assortativity is shown to be low but positive, and it has a structure
that is close to that of a human relationship. Compared to
the data network
Second, we compared the degree and the betweenness centralities of each knowledge element and found that DJs in Fig. 1 have increased degrees and betweenness centrality values. Traffic volume data frequently used for creating solutions have the highest degrees of centrality (0.033) and the highest betweenness centrality value (0.244). Additionally, the life log data has the second highest degree of centrality (0.021) and the second highest betweenness centrality (0.204) values. Based on these results, the data related to traffic and personal information act as the hub nodes in the network and appear positions that bridge different areas.
The solution with the highest degree of centrality “identified the traffic volume at the time of disaster and assessed the risks based on past disaster information, traffic volume at major points, and weak transportations, by adding traffic accident statistics.” This was accomplished based on the combination of seven traffic datasets. Conversely, the 1105 solution with the highest betweenness centrality “improved 5 the allocation efficiency by combining weather conditions, 0 latitucumdeolsoanamgjgitoaubetdreeiannrcdaetoitmtigmnieeioidonntinertvsiaoo,lsfnpsfed.iar”vtseeoaTdnddhraeasilsstabswesetehixmatasespvofiianrotcsorcmso,amtgderlapiofcvlafitieesonlrhieenndgt dbdaaotsmae,daainnosdn. tWathxeei foundd that the solutions identified with the use of data from the different areas have increased betweenness centrality values, while the solutions that combine data in the same areas have a higher degree of centrality than other solutions. The results seem quite natural, but we did not use any textual information in the description of the solutions in the analyses. It is interesting to note that the results obtained by analyzing the network of structured knowledge and their connections were in agreement with intuitive understanding based on the use of linguistic information.
In this study, we analyzed and discussed the features of the network of knowledge for data utilization. By analyzing the network of data with the contexts (solutions and requirements), we can understand the latent connections which could not be understood only with the use of data networks. In the future, we will consider the texts in knowledge elements, and we will analyze the deeper connections between data and knowledge.
Number of nodes Number of links Average degree
Link density Clustering coefficient
Assortativity Average shortest path
Diameter
Radius
This study was partially supported by JST-CREST Grant Number JPMJCR1304, and JSPS KAKENHI Grant Numbers JP16H01836 and JP16K12428.