Similar, Yet Diverse: A Recommender System

The development of Internet technologies and the concept of open source enable more and more people to be involved in generating content on the Web. Every day, an enormous amount of content is generated in various formats of audio, video and text. However, the rapidly growing volume of information makes it harder for users to identify the content they are interested in and that they can build upon. Recommender systems are one of the effective tools that can help users to cope with the information overload and can provide personalized suggestions. Typical recommender systems aiming to predict user interests are based on user profiles, demographic information and users content ratings [Barragáns-Martínez et al. 2010]. When explicit ratings are not available, implicit ratings such as users’ history of purchase and click-stream patterns become useful information for recommender systems. With the development of Web 2.0, most of these user generated content communities utilize a particular family of applications known as Social Tagging or Collaborative Tagging Systems. These applications allow users to create and share lightweight metadata in the form of chosen keywords called tags to represent the created content [Golder and Huberman 2006]. Some popular examples of these communities that support collaborative tagging are Flickr and Photobucket for photos, Last.fm and ccMixter for music and Scratch for youth game development. The tags used in these communities help users to self-organize, share and find content they are interested in [Ames and Naaman 2007]. Since tags are local descriptions of content provided voluntarily by users, they represent additional personalized information both about the user and the created content which can later be used for the creation of recommender systems [Halpin et al. 2007, Tso-Sutter et al. 2008, Liang et al. 2008]. In this paper, we propose a recommender system for the Scratch online community. Scratch users create, share and remix projects by using the Scratch programming language developed by the Lifelong Kindergarten Group at the MIT Media Lab [Resnick et al. 2009, Roque et al. 2013]. The proposed recommender system utilizes project tag information to determine similarities between various users and then uses these relationships to identify the optimal set of items to be recommended. Our aim is twofold: (i) Create and evaluate recommendations based on two different types of input tags (explicit and implicit) and (ii) evaluate recommendations based on relevancy and diversity. We are including diversity in our recommendation algorithm under the assumption that recommending a more varied set of items will be more valuable to users than simply recommending similar items. Through a calculated combination of relevancy and diversity, our recommender system is aimed at leading users to explore further into the Scratch community and improving the productivity of “passive producers” by using the output of “active consumers” [Monroy-Hernández and Resnick 2008].