Top-k Search in Product Catalogues

In the era of huge datasets, the top-k search becomes an effective way to decrease the search time of top-k objects. The original top-k search requires a monotone combination function and lists of objects ordered by attribute values. Our approach of the top-k search is motivated by complex user preferences over product catalogues. Such user preferences are composed of the local user preferences of the attributes’ values (user defined arbitrary fuzzy functions, one for each attribute) and a user defined monotone combination function. This paper compares two different approaches of the top-k search for this type of nonmonotone query. The first approach uses several B+trees, one for each attribute, and it is based on ordered lists. The second approach is new for this type of query and requires an R-tree index.