We finally highlight three more challenges for researchers in the SRS area to consider. Social Streams. Social streams, such as Twitter or the Facebook newsfeed, syndicate user activity within a social media site or a set of sites. Millions of users who share and interact in social media create a firehose of data in real-time that poses new types of challenges in terms of filtering and personalization. There are different types of streams in terms of the data they contain (homogenous as in Twitter or heterogeneous as in Facebook), the source of data (a single site or a group of sites), its access-control (public or friends-only), and subscription model (following or “friending”). As demonstrated in the Twitter-related work reviewed in this chapter, the stream’s data is different than “traditional” social media content: it represents an activity rather than an artifact or an entity; it is more intensive as one entity (e.g., a wiki page) may have a large amount of activities (e.g., edits); it may be very noisy (e.g., multiple wiki edits might not be of interest); its freshness is key: items that are few days old might already be irrelevant; and it is sparse in content and metadata (e.g., Twitter messages are limited to 140 characters). Due to all these unique characteristics, recommending social stream items become a challenge on its own within the SRS domain, and as social information continues to grow, handling this task is becoming both more challenging and more important \cite{Freyne:2010:SNF:1864708.1864766,Guy:2011:PAS:2043932.2043966,paek2010predicting}. On the other hand, the stream data can also be used to model users’ interests. Its fresh and concise nature can help build a user model that is up-to-date, identify changes in users’ tastes and preferences in real-time, and detect global trends that may influence the recommendation strategy \cite{GarciaEsparza:2010:RWS:1864708.1864773,Phelan:2009:UTR:1639714.1639794}. Beyond accuracy and evaluation over time. Many of the studies we reviewed focused on measuring the effectiveness of recommendation by their accuracy. As social recommendation proliferate, it is more important than ever to consider the bigger picture when evaluating the value of recommendation. Typical beyond-accuracy measures should be considered, including serendipity, diversity, novelty, and coverage \cite{McNee_2006}. In addition, the effectiveness of recommendation should be compared against the case where no recommendation would have been provided. Recommendations that can make the user discover and take action regarding an item s/he would not have noticed otherwise, are obviously more valuable. In many of the works we reviewed, evaluation was based on a one-time user survey. Longer term evaluation is required as the results may substantially change over time. Techniques that learn and adapt over time based on user behavior are going to be essential. Additionally, evaluation that examines the broader effect of the recommendation on the surrounding ecosystem of users, as demonstrated in \cite{Daly_2010,said2014you} is a highly desirable direction for SRS evaluation. This requires new tools and creative thinking to be brought into the existing evaluation methods. Cross-domain Analysis. As we discussed, migrating data from one social media service to another may go a long way enhancing recommendations and help deal with the cold start problem for new users. Indeed, using another site’s network, tags, and other types of information have been performed by various previous systems as mentioned in this chapter. Yet, social media sites differ in many aspects. It is not certain that one’s travel network can serve as a reliable source of recommendation for recipes. Similarly, the tags used in a news site context are not necessarily valuable for video recommendation. More research is due to explore the common and different among social media systems and when information can effectively port from one application to another to be used for recommendation. Cross-domain recommendations in RS have always been harder to explore since they require richer datasets and involve more complex use cases and research questions. As social media continues to evolve, it will be more important to explore and better understand these complexities.