In this section, we review a body of research that explored recommendation of mixed social media items within the enterprise, and included three main studies. The first study \cite{Guy:2009:PRS:1639714.1639725} focused on recommendation based on social relationships. As previously mentioned, social media enables the exposure of different types of social relationships in a way that has never been possible before. The study explored a rich set of indicators for social relationships based on social media data and compared two types of networks as basis for recommendation: familiarity and similarity. The familiarity network was built based on explicit and implicit signals from enterprise social media, such as articulated connection within an enterprise SNS, tagging one another, or co-authorship of the same wiki page. The similarity network was based on common activity in enterprise social media, such as membership in the same communities, usage of the same tags, or commenting on the same blog posts. An “overall” network was also examined, combining the two types of relationships. The recommendation score of item \(i\) to user \(u\) was determined by the following formula:
\[e^{-\alpha t(i)} \sum_{v \in{N^T(u)}} S^T[u,v] \sum_{r\in{R(v,i)}}W(r)\]
where \(t(i)\) is the number of days passed since the creation date of \(i\); \(\alpha\) is a decay factor; \(N^T(u)\) is the set of users within \(u\)’s network of type \(T\) (\(T\) \(\in\) {familiarity,similairty,overall}); \(S^T[u,v]\) is the relationship score between \(u\) and \(v\) based on the network of type \(T\); \(R(v,i)\) is the set of all relationship types between user \(v\) and item \(i\) (authorship, membership, etc.); and \(W(r[v,i])\) is the corresponding weight for the user-item relationship type between user \(v\) and item \(i\). Ultimately, the recommendation score of an item, reflecting its likelihood to be recommended to the user, may increase due to the following factors: more people within the user’s network are related to the item, stronger relationships of these people to the user, stronger relationships of these people to the item, and freshness of the item.
The recommendation widget, depicted in Figure \ref{fig:fig1}, presented the recommendations with explanations, which displayed the people who served as the “implicit recommenders” and how they were related to both the user and the recommended item. One of the key research questions of the study was whether explanations have influence on the instant interest in the recommended items. This was examined by comparing recommendations with and without explanations.