SOCIAL COMPUTING Algorithm for Implementing Our Framework calculate the source qualification for user u, dation is relative to their qualification. The recom- Quk, as a weighted average of various indica- mendation function of an item i to a user u could be tors. A simple formula is based on various algorithms, the gold standard in CF systems being Quk= WH x Hu k+ Wrx Tuk+ Wrs x TSu k+WR X Ruk where Huk is the homophily(shared preferences)score P,,=r,+ users u and k, Tu,k is the trust score, TSu, k is the tie y strength(interaction frequency) score, and Ruk is the reputation score. W represents the relative weight where Pu i is the prediction score of item i to user u, r assigned to each indicator: WH for homophily, W for is the average overall past ratings provided by user u, Alternative formulas, such as harmonic mean, are the average overall past ratings provided by user( trust, Wrs for tie strength, and We for reputation rk.i is the rating assigned to item i by user u, and also possible. The system prediction component's output is a prediction of item relevancy to users. The Reference the effect of each of n sources on the final recommen- vol 22, no. 1, 2004, pp 5-55 >ans h ive Filtering system computes it as an aggregation of the recom- 1. J. Herlocker et al., "Evaluating Collaborative Filtering mendations of the n most qualified sources, where about how to incorporate various indicators of based on efficiency considerations. The impact social relationships into recommender systems. of the various indicators on system efficiency is We grounded our proposed framework on be- independent of task domain. Efficiency depends havioral theory, utilizing a series of relationship on three key factors indicators that we can extract in online settings We expect this framework to provide accu- effort required by users racy enhancements beyond traditional CE, es- effort required by system administrators, and ecially in cold-start situations. This problem is privacy concerns. critical in commercial recommender systems+ 12 because in the early phases of CF system deploy- Table 2 summarizes these considerations for the ment, relatively little information on user tastes various relationship indicators is available, making it difficult to provide accu- The effort required from users plays a large rate recommendations. For example, two of role in determining system adoption. To keep the most popular commercial CF applications- user effort down to a minimum, the system can GroupLens and Epinions-suffer from the cold- calculate shared preferences based on users start proble consumption records. It can also capture and Advice-taking literature suggests that relation- calculate e communicat ation frequency automati ship indicators such as trust and tie strength cally. Establishing a social network(whether to are highly correlated with homophily. It makes calculate trust or indicate reputation) requires ense,then, that data extracted from a social net- users to invite and accept invitations from other work could serve as a proxy for preference simi- users, whereas a reputation system requires them larities in cold-start situations and ensure that to rate the quality of the recommendations the the system associates a recipient with appropriate received sources The effort required from system administra- tors, too, might play a part in decisions about Effort and Privacy which relationship indicators to use. Calculating Our proposed framework is somewhat generic in shared preferences requires the recording of user the sense that it includes all available relationship profiles-and matching them. Although calculat indicators. However, any implementation of this ing direct trust relationships from a given social framework is likely to use a subset of indicators. network is straightforward, propagating trust to We can choose which indicators to use based on indirect relationships requires additional calcu the domain in which we deploy the system and lations. We can calculate reputation scores from TPr。 July/August200942 IT Pro July/August 2009 Social Computing about how to incorporate various indicators of social relationships into recommender systems. We grounded our proposed framework on be￾havioral theory, utilizing a series of relationship indicators that we can extract in online settings. We expect this framework to provide accu￾racy enhancements beyond traditional CF, es￾pecially in cold-start situations. This problem is critical in commercial recommender systems4,12 because in the early phases of CF system deploy￾ment, relatively little information on user tastes is available, making it difficult to provide accu￾rate recommendations.13 For example, two of the most popular commercial CF applications— GroupLens and Epinions—suffer from the cold￾start problem.10,13 Advice-taking literature suggests that relation￾ship indicators such as trust and tie strength are highly correlated with homophily. It makes sense, then, that data extracted from a social net￾work could serve as a proxy for preference simi￾larities in cold-start situations and ensure that the system associates a recipient with appropriate sources. Effort and Privacy Our proposed framework is somewhat generic in the sense that it includes all available relationship indicators. However, any implementation of this framework is likely to use a subset of indicators. We can choose which indicators to use based on the domain in which we deploy the system and based on efficiency considerations. The impact of the various indicators on system efficiency is independent of task domain. Efficiency depends on three key factors: • effort required by users, • effort required by system administrators, and • privacy concerns. Table 2 summarizes these considerations for the various relationship indicators. The effort required from users plays a large role in determining system adoption. To keep user effort down to a minimum, the system can calculate shared preferences based on users’ consumption records. It can also capture and calculate communication frequency automati￾cally. Establishing a social network (whether to calculate trust or indicate reputation) requires users to invite and accept invitations from other users, whereas a reputation system requires them to rate the quality of the recommendations they received. The effort required from system administra￾tors, too, might play a part in decisions about which relationship indicators to use. Calculating shared preferences requires the recording of user profiles—and matching them. Although calculat￾ing direct trust relationships from a given social network is straightforward, propagating trust to indirect relationships requires additional calcu￾lations. We can calculate reputation scores from Algorithm for Implementing Our Framework We calculate the source qualification for user u, Qu,k, as a weighted average of various indica￾tors. A simple formula is Qu,k = WH × Hu,k + WT × Tu,k + WTS × TSu,k + WR × Ru,k, where Hu,k is the homophily (shared preferences) score for users u and k, Tu,k is the trust score, TSu,k is the tie strength (interaction frequency) score, and Ru,k is the reputation score. W represents the relative weight assigned to each indicator: WH for homophily, WT for trust, WTS for tie strength, and WR for reputation. Alternative formulas, such as harmonic mean, are also possible. The system prediction component’s output is a prediction of item relevancy to users. The system computes it as an aggregation of the recom￾mendations of the n most qualified sources, where the effect of each of n sources on the final recommen￾dation is relative to their qualification. The recom￾mendation function of an item i to a user u could be based on various algorithms, the gold standard1 in CF systems being , where Pu,i is the prediction score of item i to user u, is the average overall past ratings provided by user u, rk,i is the rating assigned to item i by user u, and is the average overall past ratings provided by user k. Reference 1. J. Herlocker et al., “Evaluating Collaborative Filtering Recommender Systems,” ACM Trans. Information Systems, vol. 22, no. 1, 2004, pp. 5–53