This paper presents item-based top-N recommendation algorithms that are significantly faster and provide comparable or better quality recommendations than traditional user-based collaborative filtering systems. The proposed algorithms determine the similarities between items and use these similarities to identify the set of items to be recommended. The key steps in these algorithms are (i) the method used to compute item similarities and (ii) the method used to combine these similarities to compute the similarity between a basket of items and a candidate recommender item. Two methods for computing item-to-item similarity are presented: one based on cosine similarity and another based on conditional probability. The cosine-based method treats items as vectors in the user space and uses the cosine measure to compute similarity. The conditional probability-based method computes item-to-item similarities using a technique inspired by the conditional probability between two items, extended to differentiate between users with varying amounts of historical information and between frequently and infrequently purchased items. A method of combining these similarities is presented that accounts for item-neighborhoods of different density, which can incorrectly bias the overall recommendation. Finally, a class of higher-order item-based algorithms is presented that obtain the final recommendations by exploiting relations between sets of items. The algorithms were experimentally evaluated on nine real datasets and 36 synthetic datasets. The results show that item-based algorithms are up to two orders of magnitude faster than traditional user-based systems while achieving comparable or substantially higher quality. The experiments also show that similarity normalization and row normalization significantly improve the performance of the algorithms. The results indicate that item-based algorithms are effective in providing accurate and efficient recommendations, especially for large-scale applications. The paper concludes that item-based recommendation algorithms are a promising alternative to user-based collaborative filtering for top-N recommendation tasks.This paper presents item-based top-N recommendation algorithms that are significantly faster and provide comparable or better quality recommendations than traditional user-based collaborative filtering systems. The proposed algorithms determine the similarities between items and use these similarities to identify the set of items to be recommended. The key steps in these algorithms are (i) the method used to compute item similarities and (ii) the method used to combine these similarities to compute the similarity between a basket of items and a candidate recommender item. Two methods for computing item-to-item similarity are presented: one based on cosine similarity and another based on conditional probability. The cosine-based method treats items as vectors in the user space and uses the cosine measure to compute similarity. The conditional probability-based method computes item-to-item similarities using a technique inspired by the conditional probability between two items, extended to differentiate between users with varying amounts of historical information and between frequently and infrequently purchased items. A method of combining these similarities is presented that accounts for item-neighborhoods of different density, which can incorrectly bias the overall recommendation. Finally, a class of higher-order item-based algorithms is presented that obtain the final recommendations by exploiting relations between sets of items. The algorithms were experimentally evaluated on nine real datasets and 36 synthetic datasets. The results show that item-based algorithms are up to two orders of magnitude faster than traditional user-based systems while achieving comparable or substantially higher quality. The experiments also show that similarity normalization and row normalization significantly improve the performance of the algorithms. The results indicate that item-based algorithms are effective in providing accurate and efficient recommendations, especially for large-scale applications. The paper concludes that item-based recommendation algorithms are a promising alternative to user-based collaborative filtering for top-N recommendation tasks.