We are excited to announce a new recipe in Amazon Personalize that, when given an item, will recommend similar items based on both user-item interaction data and item metadata. The combination of your users’ historical interactions and the information you have about your items increases the relevance of recommendations and ensures similar items capture your users’ attention. To assess the similarity of items we measure how frequently the items are found together in users’ histories. As a benchmark, we found that the new recipe is 10.2% more accurate in identifying similar items than recipes that use interactions data alone. This means your users will be more likely to find the items most related to what they are viewing.

Categories: AWS