Avoiding monotony

摘要: The primary premise upon which top-N recommender systems operate is that similar users are likely to have tastes with regard their product choices. For this reason, algorithms depend deeply on similarity metrics build the recommendation lists for end-users.However, it has been noted products offered often too each other and attention paid towards goal of improving diversity avoid monotonous recommendations.Noting retrieval a set items matching user query common problem across many applications information retrieval, we model competing goals maximizing retrieved list while maintaining adequate as binary optimization problem. We explore solution strategy by relaxing trust-region problem.This leads parameterized eigenvalue whose finally quantized required solution. apply approach prediction problem, evaluate system performance Movielens dataset compare standard item-based algorithm. A new evaluation metric ItemNovelty proposed in work. Improvements both accuracy obtained compared benchmark