*In: Networked & Electronic Media Summit 2009 (NEMS2009, St-Malo). Current details for correspondence: marrow@mail.upb.de MyMedia: Producing an Extensible Framework for Recommendation Paul Marrow 1* , Rich Hanbidge 2 , Steffen Rendle 3 , Christian Wartena 4 , Christoph Freudenthaler 5 1 BT Innovate & Design, Ipswich, UK; 2 EMIC, Aachen, Germany; 3 University of Hildesheim, Hildesheim, Germany; 4 Novay, Enschede, The Netherlands; 5 University of Hildesheim, Hildesheim, Germany 1 paul.marrow@bt.com, 2 richhan@microsoft.com, 3 srendle@ismll.de, 4 christian.wartena@novay.nl, 5 freudenthaler@ismll.de Abstract: Users and implementers of multimedia today face a common problem: how to deal with the “crisis of choice” that exists when very many different forms of multimedia are presented to users. In such circumstances search is not a complete solution, recommendation can improve the user experience. However there are few recommender system solutions that are sufficiently versatile. This paper outlines the MyMedia software toolkit which is the outcome of an international collaboration to develop an extensible software framework for multimedia recommendation, incorporating cutting edge recommender algorithms, metadata enrichment, and software design. It will be tested in field trials under realistic conditions and has been made available to the research community as open source software. Keywords: Multimedia, Software, Framework, Metadata Enrichment, Recommender, Field Trial, Open Source. 1 INTRODUCTION Users and developers of multimedia applications at the present time are faced with a common problem: the need to process and present many different formats of multimedia. For the multimedia processing and management software developer, this means requirements to make software as flexible and extensible as possible, requirements which may be difficult to achieve when near- or long-term changes in multimedia trends are difficult to predict. For the user, this means being faced with a “crisis of choice” when being presented with many different forms of media and many different means of interacting with them. Search provides one means to assist users towards reaching desired multimedia items, but can be of limited use if they are unclear about their aims. This is where recommendation can be useful. Here we introduce the MyMedia toolkit, which provides a framework for the input of multimedia content and associated metadata, the processing and enrichment of that metadata, and the passage of content, enriched metadata and user information to recommender algorithms in order to generate recommendations in support of user requirements. In effect it is a dynamic personalisation framework that links the different components required for effective recommendation with user recommendation and feedback. The software is designed to provide an intuitive abstraction of personalization components, while maintaining flexibility and performance. This allows developers from many areas of expertise to contribute substantially to the personalization scenarios. Also, the framework allows for full application development; from rapid prototyping to field trial design and execution. The end result is a pluggable and extensible software toolkit. The following sections describe the MyMedia toolkit in more detail and its proposed uses. Section 2 discusses related work in recommender algorithms and systems. Section 3 describes the core software framework which underpins the functionality of the MyMedia framework. Sections 4 and 5 deal with significant components of the MyMedia software framework: the recommender algorithms, and the metadata enrichment software. Finally we discuss the proposed uses of the software, in the near future in field trials, and the current open source release. 2 RELATED WORK Recommender algorithms have been an active area of research since the mid-1990s, e.g. [16][17]. The field has diversified to study algorithms that draw upon different sources of information to produce recommendations to users [1]: methods based on content, methods based on user profiling, especially collaborative filtering which builds upon similar preferences of different users, and hybrid methods that bring together different types of recommender algorithms [4]. Specific applications of recommender algorithms have been developed in many areas, such as e-commerce, with amazon.com and its subsidiaries recommending books, CDs and other goods [2], MovieLens [13] recommending movies, and last.fm [9] recommending music, for example. What is less common is the development of frameworks for recommendation that will allow flexibility in multiple areas, especially: (a) the type of recommender algorithm used; (b) the format of content used; (c) the format of metadata associated with content accepted; (d) the type of metadata enrichment associated with the recommender algorithm used (if appropriate) and (e) the recommender application and user interface made available to the user.