Article Preview
Top1. Introduction
Nowadays, Online Social Networks (OSNs) represent the most natural environment that allow users creating and sharing multimedia contents such as text, image, video, audio for different purposes (e.g., comment events and facts, declare and share personal opinions about a specific topic, share moments of their life etc.). Thus, millions of individuals can create online profiles and share personal information within more and more vast networks of people.
Indeed, by means of shared social media content each user can “indirectly” interacts with the others generating particular “social links” that can effectively characterize their behaviors within the network and can support a lot of Social Network Analysis (SNA) applications. In such a context, multimedia data can play a key-role: specifically, representing and understanding user-multimedia interaction mechanisms and multimedia items’ characteristics can be useful to predict user behavior and, especially, to design human-centric multimedia services.
With the exponential growth of social media, it is quite important to provide multimedia information of real interest for users: which photo to watch in Flickr, which music to listen in Last.Fm, which video to watch in YouTube, etc., just to provide several examples.
Thus, Recommender Systems (Kantor, 2015) surely represent one of the most important tool that can be needed within OSNs, due to their capability of providing personalized and useful contents to users on the basis of their needs and preferences. As an example, they have been used in the last dedecade to support users in the following tasks: what items to buy (Kazienko & Kolodziejski, 2006), which photo or movie to watch (Albanese, d’Acierno, Moscato, Persia & Picariello, 2013), (Lekakos & Caravelas, 2008), which music to listen (Yoshii, Goto, Komatani, Ogata & Okuno, 2008), what travels to do (Colace, De Santo, Greco, Moscato & Picariello, 2015), or even who they can invite to their social network (Stan, Muhlenbach & Largeron, 2014), which artwork could be interesting within an art collection or even to suggest visiting paths in Cultural Heritage applications (Albanese, d'Acierno, Moscato, Persia & Picariello, 2011), (Bartolini, Moscato, Pensa, Penta, Picariello, Sansone & Sapino, 2016).
However, social media recommendation is quite different from traditional recommendation approaches because it needs to take into account not only content information and users’ similarities (as in the most diffused recommender systems), but also users’ social relationships and behavior within an OSN to handle a large amount of multimedia contents showing Big Data features, mainly due to their high change rate, their huge volume and intrinsic heterogeneity.
In this context, one of the most interesting open research challenge is to provide recommendation techniques for multimedia data in one or more social environments, exploiting at the same time (low-level) features and (high-level) metadata description (together with the attached semantics) of contents together with users’ community behaviors in the different OSNs, and eventually considering the context information (Amato, Moscato, Picariello & Sperlí, 2017), (Kabassi, 2013) as a further criterion to have more accurate results in the recommendation process.