Filter-Based Product Search Engines With Dynamic Component Ranking

Abstract

The use of faceted browsing is common on shopping and comparison websites. When dealing with problems of this kind, it is usual practise to apply a specified set of features in a certain order. This tactic suffers from two major flaws that undermine its effectiveness. First things first: before you do anything else, you need to make sure that you set aside a significant amount of time to compile an effective list. Second, if you have a certain number of aspects and all of the products that are relevant to your search are tagged with the same aspect, then that particular aspect is basically worthless. This article presents a method for doing online business that makes use of a dynamic facet ordering system. On the basis of measurements for specificity and dispersion of aspect value dispersion, the entirely automated system assigns ratings to the characteristics and facets that lead to a speedy drill-down for each and every prospective target product. In contrast to the methodologies that are currently in use, the framework takes into consideration the subtleties that are specific to e-commerce. These nuances include the need for several clicks, the grouping of facets according to the traits that they share, and the predominance of numerical facets. In a large-scale simulation and user survey, our approach performed much better than the baseline greedy strategy, the facet list prepared by domain experts, and the state-of-the-art entropy-based solution. These comparisons were made using the same data.