Collaborative Filtering Model for Subscriber Offer Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting subscriber preferences in service providers are limited, as they only allow independent matching of subscribers to offers without enabling predictions among offers, leading to reduced accuracy.
Innovation Solution
A collaborative filtering system and method that calculates ranked predictions for subscribers by correlating subscriber and offer data, using a collaborative filtering algorithm to generate a model for predicting implicit rankings of new correlations between subscribers and offers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate models are created for each offer using logistical regression algorithms, then the system can calculate the probability of subscriber acceptance for individual offers, but the system cannot make predictions among offers themselves to determine which offer a subscriber is most likely to accept
Solution Approach 1:
The patent merges multiple separate logistical regression models into a unified collaborative filtering framework. Instead of treating each offer independently, the system combines offer data into a unified structure where offers are compared relative to each other, enabling both individual offer prediction and inter-offer ranking simultaneously.
Solution Approach 2:
The patent introduces a new dimension of analysis by calculating implicit rankings that order offers relative to each other. This transforms the prediction problem from a single-dimensional binary classification (accept/reject) into a multi-dimensional framework that includes both acceptance probability and relative offer ranking, allowing the system to identify the most preferred offer among multiple options.
2Ease of operation
If independent matching is used for each subscriber-offer pair, then the system can process offers individually, but the system lacks the ability to rank and compare multiple offers to predict which one a subscriber will accept
Solution Approach 1:
The patent performs preliminary calculations of implicit rankings for all offer pairs before making final predictions. By pre-computing the relative ordering of offers and storing these rankings, the system simplifies the final prediction step while maintaining high accuracy, as the complex inter-offer comparisons are resolved in advance.
Solution Approach 2:
The patent introduces implicit rankings as an intermediary structure that mediates between simple independent offer processing and complex inter-offer comparisons. This intermediary layer allows the system to maintain processing simplicity while achieving accurate predictions, as the implicit rankings serve as a pre-processed reference that simplifies final offer selection.
Data Source
AI summary
As described herein, a collaborative filtering system, method, and computer program provide a model used to calculate ranked predictions for subscribers. In use, data is received which correlates subscribers of a service provider with offers provided to the subscribers by the service provider, wherein each correlation is associated in the data with an indication of whether the offer was accepted by the subscriber. From the data, a plurality of unique correlations between subscribers and offers are determined. An implicit ranking is calculated for each unique correlation, based on the associated indication of acceptance for correlations matching the unique correlation. Input data is generated that includes each unique correlation with the implicit ranking calculated for the unique correlation. The input data is processed utilizing a collaborative filtering algorithm to generate a model configured to provide predicted implicit rankings for new correlations between subscribers and offers.