HyperRank is a user-user collaborative filtering algorithm that I have designed for use as a simple recommendation engine and have implemented here in Python.
Mathematically can be written as:
where
Given a set of users and items, represented as user_data and item_data respectively, we can calculate a numerical ranking score for each item using the HyperRank algorithm.
The HyperRank algorithm involves the following steps: -Assign an integer value to each item based on the number of votes it has received from users. -Rank the items based on their integer values, with higher values indicating a higher rank. -Divide the ranked items into three tiers: A, B, and C. -Apply a scaling factor to the integer values of items in tiers B and C, with the scaling factor increasing for each tier. -Re-rank the items based on their updated integer values. -Once the items have been ranked, we can recommend them to a current user based on their past votes. -The HyperRank class provides methods to perform each of these steps, and the hyper_rank method combines them to generate a ranked list of items that the current user may be interested in.
Item x is a member of set {T}, which has been partitioned into three subsets {TA, TB, TC} based on the popularity of items within the entire set. Items in {TA} remain unchanged, items in {TB} are squared, and items in {TC} are cubed. {TA} represents the top 25% of most popular items, {TB} represents items between the 25th and 75th percentiles, and {TC} represents the bottom 25%. The purpose of partitioning {T} into three subsets is to provide greater weight to items that are of interest to two users but not generally popular in the set as a whole. This metric significantly increases the similarity between two users when the function is executed.