How can we improve the acceptance of recommendations in collaborative systems? The group identity of recommenders and recipient involvement in group formation impacts on the likelihood that users of social collaborative systems would accept recommendations provided on it. We introduce the term 'friends group' to describe a sub-group of the 'neighbors group' in recommender systems that is not solely rank-dependent, as opposed to 'neighbors' that are assigned by rating similarity. The 'friends group' is unique because of the user's involvement in its formation and the user's ability to choose the characteristics of its members. The latter aspect corresponds to Festinger's "Social Comparison Theory", suggesting that 'neighbors' (like-minded groups) are relevant for 'low-risk' domains whereas similarity-based 'friends' are more relevant for 'high-risk' domains. We conducted a two year field study, using QSIA, a Web-based Java-programmed collaborative system for collection, management, sharing and assignment of learning knowledge items. QSIA was implemented in over ten courses in several universities. QSIA database and logs contained approximately 31,000 records of items-seeking acts, 3,000 users, 10,000 items, 3,000 rankings and knowledge items from 30 domains. We found that the difference between acceptance and rejection ratios of recommendations when the items originated from an advising group comprised of 'friends', is significantly higher than when the advising group is the more commonly known 'neighbors group'. The difference increases for frequently recommended as opposed to other items and for experienced as opposed to 'average' users. Our longitudinal analysis indicates a positive learning curve for experienced users, who, over time, increasingly preferred 'friends group' over 'neighbors group' as their experience with the system increases. Also, users chose their own group to participate in the advising group significantly more than other groups. The contribution of this study is in explicating the relationship between the perceived quality of the recommendation (measured in terms of "usage actions"), and the user's involvement in the formation of the advising group. The major implication of our findings for the development of recommender systems is the need to enhance involvement of recommendation seekers in the process of forming the advising group. Developers of recommender systems should consider increasing users' control over relevant characteristics of the members of this group.