User Centric Similarity Search

Abstract

In market analysis, user priorities are very important. There has been a massive amount of work on query primitives in the database literature, such as the well-known top-k query, which can be used to rank products based on consumer preferences. Nonetheless, the basic operation of evaluating product similarity is often performed without regard for these preferences.Instead, products are represented in a feature space based on their characteristics, and similarity is calculated using conventional distance metrics. In this paper, we use product rankings based on consumer feedback to map products in a user-centric space where similarity calculations are performed. We define some of the mapping's most significant characteristicsthat result in upper and lower similarity boundsin order to perform these user-centric similarity computations, we can use traditional multidimensional indexes on the original product space. We demonstrate how to efficiently perform interesting similarity calculations inspired by the widely used range and nearest neighbor queries, thus pruning significant portions of the databased on the bounds we derive on the user-centric similarity of products.