Nowadays the amount of content and services available on the Internet is awesome.
This availability has had the merit to greatly enrich the multimedia user experience. Nevertheless, it has made the ability to match his/her needs more and more complex.
Consequently, in recent years market-leading search engines have been augmented with “recommendation systems”.
By interacting with a given application, a specific user can receive tailored contents/services thus experiencing an enhanced fruition.
Nevertheless, if the user is not satisfied, he/she is forced to open a different application which will possibly recommend different contents/services.
By moving from an application to another one, no information about historical activity of the user are exploited: each application “vertically” provides its recommendations.
Consequently, the user may experience a fragmented and deteriorated fruition.
The MPEG- 21 “User Description” (MPEG-21 UD) standard formalizes interoperability between different recommendation systems.
By exploiting the availability of standard descriptions, a given application can integrate “horizontally” recommendations from different sources, thus making more fluid and satisfactory the user experience.
Nowadays, the availability of large-scale data poses serious limitations in terms of usability.
According to Kosner (1), in the last few years, an overwhelming growth in the number of services and multimedia contents accessible on the Web has been observed.
Together, the amount of smart devices and Internet users has reached an impressive size.
A recent study by Cisco (2) argues that, by 2019, 5.2 billion global mobile users and 11.5 billion mobile-ready devices and connections will be foreseen.
In order to better match user needs and improve his/her experience, recommender systems can be adopted.
By suggesting potentially interesting or useful items to users, these systems are aimed at addressing the information overload problem.
Nevertheless, the accuracy of recommendations a user receives might strongly affect the quality of his/her experience. During an ordinary session, a generic user likely enjoys several contents and services.
These might belong to different “domains”, namely different items (e.g. “movies”, “books” etc.) or genres (e.g. “action”, “comedy”, etc.). In addition, these contents might be provided by vertical and closed1 systems, e.g. different departments of the same provider, or, by independent providers.
So, in the attempt of satisfying user needs, these stakeholders (internal or external to a given company), “vertically” return to him/her distinct recommendations, i.e. recommendations not mutually related and restricted to specific and separated domains (e.g. Netflix suggests movies or TV series, Youtube suggests music songs, etc.).
This vertical approach might impact on the overall quality of the user experience.
It is clear that alternative solutions to these issues are needed: on one hand, new business models (which, to date, have been lacking) focused on “horizontal” agreements between different stakeholders are expected; on the other hand, exchanging standard information among these stakeholders should be taken into account.
Regarding the opportunity of testing new business models, two recent initiatives face the journalism challenges: Instant Articles² and Digital News Initiative³.
In the former, Facebook aims to “transform the way users read news articles” and has signed up different media companies and publishers (BuzzFeed, the New York Times, National Geographic, NBC News and The Atlantic, BBC News, the Guardian, Bild and Spiegel).
In the latter, “Google will work hand in hand with news publishers and journalism organizations (Les Echos, FAZ, The Financial Times, The Guardian, NRC Group, El Pais, La Stampa, Die Zeit) to help develop more sustainable models for news”.
Both initiatives are aimed at merging services provided by multiple publishers.
Nevertheless, in any of these initiatives there is an explicit task to ensure recommendations across different domains and/or partners.
In the last few years, cross-domain recommendation algorithms have been proposed to address the cold-start issue, thus improving accuracy, offering serendipity, and enhancing user models.
See for example the works by Deng et al (3), Tobiias et al (4), Abel et al (5).
A generic mediation mechanism for integrating user modelling data was proposed by Berkovsky et al (6).
The mediation mechanism is aimed at facilitating interoperability between recommender systems thus providing more complete and usable recommendations to the users.
Bringing this approach further, the MPEG-21 User Description standard (MPEG-21 UD) (7) aims at ensuring the interoperability between recommendation engines of different stakeholders thus enhancing recommending accuracy and improving the user experience.
MPEG-21 UD does not standardize the way in which recommendation engines work, i.e. algorithms adopted to generate the recommendations (e.g., collaborative and/or content- based filtering).
Rather, it defines standard descriptions about a given user, his/her context, the services and/or the items he/she has enjoyed and other information that can be relevant for the purpose of recommendation.
Additionally, by defining standard descriptions of the recommended items it allows for a more efficient cross-domain and cross-service integration of recommendations.
The rest of the paper is organized as follows: we first outline the basics of MPEG-21 UD standard; then we briefly describe two reference use cases, namely a Web based service involving news recommendations, and a personalized hybrid digital media service; finally we give some examples of standard descriptors before some concluding remarks.