My Health Dictionary: Study on Web Service using Program Information Data-hub as Linked Open Data Masaru Miyazaki1 , Makoto Urakawa1 , Ichiro Yamada1 , Kikuka Miura1 , Taro Miyazaki1 , Hiroshi Fujisawa1 , and Toshio Nakagawa1 Japan Broadcsting Corporation(NHK), Tokyo, Japan, miyazaki.m-fk@nhk.or.jp Abstract. With the evolution of the global Internet, it has become increasingly common for companies to automatically exchange various types of data among themselves. In addition, content providers, such as broadcasting stations, are being required to change their content-serving strategy so that the content can be delivered to viewers via various external services. To address this strategy change, this paper proposes program-related information as machine-readable web data that can be used in internal and external services. We report on the construction of a program information data-hub using the Linked Open Data (LOD) stan- dard format recommended by the World Wide Web Consortium. Results obtained by prototyping the data-hub and associated web services show that services employing a variety of program information can be realized by representing knowledge about the content as LOD data. Keywords: linked open data, program information, health 1 Introduction The circulation of large amounts of content on the Internet, the rapid spread of mobile devices such as smartphones, and the appearance of new viewing styles such as time-shift playback have resulted in a significant change in the behavior of viewers of TV programs. In line with this change, content providers such as broadcasting stations are being required to change their content-serving strategy, so that the content can be delivered to viewers via various external services rather than simply waiting for them to access content via a broadcast and video on demand service. We believe that an effective strategy to achieve this objective is to utilize semantic web and Linked Open Data (LOD) technology to build a “Web of Data.” Consequently, we are currently studying how to build a hub comprising various types of data associated with a TV program by creating LOD program-related information and external data. In this demonstration, we give an outline of the program information data-hub and introduce the prototype health information services that use the data-hub. 2 Related Works The BBC recognized the possibility of improving the utilization and presence of content via LOD technologies at an early stage, and they have been working to build a content data space using LOD. They have consequently developed LOD for content such as program episodes and music artist information, and these can be referenced from a variety of external sources. Moreover, at the web site Wildlife1 , animal species and behavior information are systemized as RDF. Thus, a user can immediately enjoy relevant information and programs. Further, their efforts to connect various pieces of program information as LOD by crowdsourcing using tags are highly appreciated in the academic field[1]. In this study, our aim is to actualize a more advanced data space that can be utilized as an internal or external service of broadcasting stations. As a result, we have constructed a program information data-hub that has accumulated not only existing program information but also knowledge about the program contents, and are currently exploring its possibilities. 3 Development of program information data-hub NHK, Japan’s public broadcast station, has developed a variety of program- related information in an in-house database for the purpose of providing broad- casting information on their web site. However, much of the information was described in RDB tabular form and, as a result, was underutilized for collabo- ration with other new services. For example, a vegetable that is introduced as a disease prevention measure in a health program could also be introduced as an ingredient for a recipe in a cooking program. In this way, programs often provide information that share common concepts with another program. By connecting such programs to each other via the common concept, the creation of a new ser- vice that connects programs transversely is possible. With the aim of realizing such services, we gathered the program-related information (such as location of image, video, web site) residing in in-house databases, transformed information to RDF and automatically constructed an RDF store called a program informa- tion data-hub. We used the Programmes Ontology of the BBC as a reference in describing the schema of our data-hub, and expanded it to be able to describe NHK’s own program information. Next, in order to realize the cooperation be- tween the various external services, we automatically extracted performer names and important words included in the program information, and added the link information to the vocabulary of the DBpedia Japanese2 . Further as external knowledge, we automatically added a link to a “knowledge map,” which is a program-related knowledge dataset that is currently being built. The knowledge map comprises two types of data. One is “concept map,” which consists of data obtained by analyzing a large text corpus on the web. The map shows the semantic relations between words, such as causal relations 1 http://www.bbc.co.uk/nature/wildlife 2 http://ja.dbpedia.org and hyponymy relations[2][3]. The second is “content map,” which was generated by analyzing the summary text of a program. This method is composed of two processes: topic extraction and relation estimation based on TF-IDF statistics and semantic relation in concept map. It shows the relation between words and the associated program. A structural example of a knowledge map associated with the program infor- mation is shown in Fig.1. Concept map included 28 types, approximately 1,012 million word relations in total. Finally, we accumulated all schemas, instances, and knowledge map data in an RDF Store, and constructed an environment that is accessible from a variety of services through the SPARQL endpoints and WebAPI. Currently, program information data-hub contains 1.89 million pieces of RDF triple created from the experimental accumulation of 6,700 pieces of program data over a period of two months. Concept  Map ConceptWord relation ConceptWord “Vinegar” “Prevention” “Hypertension” Content  Map TV  Program relation URI:nhkdwc0112 “Vinegar” “topic” “Dining  with  the  chef”   No.112  “Vinegar-‐‑‒marinated  Aji  Salad” ConceptWord Fig. 1. Structure example of program information and knowledge map 4 My Health Dictionary service using program information data-hub We have developed a “My Health Dictionary”(Fig.2) as a service that utilizes the program information data-hub. My Health Dictionary is implemented as an extension of Google Chrome. When the user selects a keyword in which he/she is interested while browsing the Web, the program associated with the word is displayed as a popup. Fig.2 shows an example in which the keyword “hyper- tension” is selected on the text of a Webpage. When the user right-clicks on the selected keyword, a knob describing the keyword string is displayed, and related words linking to the keyword are displayed around the knob. If the user then selects a related word such as “prevention,” a list of programs related to “prevention of hypertension.” is then displayed. By clicking the program, the user can then watch the video of the program or browse the program website. Using the information from the concept maps, the system is also able to list the cooking program that introduced a recipe using “vinegar” which is said to help in the prevention of hypertension. Because the data-hub stores program infor- mation about various genres, the cross-sectoral services that connect the various Relation word “prevention” List of program Selected word The program about the "mineral" that helps in Knob of “hypertension” the "prevention" of ”hypertension." Fig. 2. Screen shot of “My Health Dictionary” programs can be realized by utilizing program-related knowledge such as from the concept map or the content map. 5 Conclusion and future work In this paper, we reported on a prototype LOD based program information data- hub constructed by linking external knowledge with existing program informa- tion. Further, we demonstrated a service example using the data-hub. In future work, we plan to conduct studies on data-hub construction and utilization in areas such as news, culture, and education, and establish an improved and so- phisticated data-hub for actual service operation. References 1. Yves Raimond, Tristan Ferne, Michael Smethurst, Gareth Adams: The BBC World Service Archive prototype, Web Semantics: Science, Services and Agents on the World Wide Web, Vol.27–28, pp.2–9 (2014) 2. Stijn De Saeger, Kentaro Torisawa, Jun’ichi Kazama, Kow Kuroda and Masaki Murata: Large Scale Relation Acquisition using Class Dependent Patterns, In Proceedings of the IEEE International Conference on Data Mining (ICDM’09)， pp.764–769(2009) 3. Stijn De Saeger, Kentaro Torisawa, Masaaki Tsuchida, Jun ichi Kazama, Chikara Hashimoto, Ichiro Yamada, Jong-Hoon Oh, Istvan Varga, Yulan Yan,Relation Ac- quisition using Word Classes and Partial Patterns,In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP 2011), pp.825- 835(2011)