Optimizing the Number of Neighbors in Trust Based Recommender Systems

Users in trust based recommender systems seek recommendations from their directly trusted parties known as neighbors or from neighbors of neighbors and so on. This paper proposes an appropriate range ([minrec... maxrec]) providing minimum and maximum number of recommenders that one should have in his close communication. More neighbors than the maximum number of neighbors (maxrec) suggested by the range does not improve the quality of recommendations but requires more time and computations in accumulation of extra recommendations. Lesser number of neighbors than minrec may compromise the quality of recommendations thus requiring seeking recommendations transitively. This again involves time and computation in propagating the query through chain of neighbors and getting the responses. A method to maintain number of neighbors within this range is also proposed. Experiments were conducted on real datasets to discover the most appropriate number of neighbors that an agent should keep.