Recommendation Score Optimization via Uncertainty Correction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing content recommendation systems rely on predicted scores with inherent prediction errors, failing to optimize recommendations to meet individual user needs, as they equally risk overrating or underrating content interest, leading to unsatisfactory results for users who prioritize avoiding uninteresting items or maximizing interesting ones.

Innovation Solution

A system and method that corrects predicted recommendation scores using an uncertainty factor and a correction coefficient to optimize scores, allowing for personalized recommendations by adjusting the impact of uncertainty based on user preferences and content types, thereby enhancing user satisfaction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If standard recommender systems use predicted recommendation scores with Gaussian distribution of prediction errors, then the system can provide recommendations based on highest predicted scores, but the system cannot optimize recommendations to meet individual user needs for avoiding uninteresting items or maximizing interesting ones

Engineering Contradiction:
ImproveAbility to optimize recommendations according to user preferencesVSAvoidPrediction accuracy of recommendation scores
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent applies local quality by differentiating the treatment of prediction errors based on their sign. Instead of treating all prediction errors uniformly, the system applies asymmetric penalty functions: positive prediction errors (overestimations) are penalized differently than negative prediction errors (underestimations). This allows the system to locally adapt to user preferences by adjusting the penalty weight for overestimations versus underestimations, thereby optimizing recommendations according to whether users prioritize avoiding uninteresting items or maximizing interesting ones.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes the parameter of the penalty function from a symmetric to an asymmetric form. By introducing a parameter that controls the relative weight of positive versus negative prediction errors, the system can dynamically adjust the recommendation optimization strategy. This parameter change enables the system to flexibly adapt to different user preferences without sacrificing prediction accuracy, resolving the contradiction between adaptability and measurement precision.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If the system minimizes root mean square error of prediction scores, then overall prediction accuracy is improved, but users still receive both overrated and underrated items with equal probability

Engineering Contradiction:
ImproveRoot mean square error minimizationVSAvoidUser satisfaction with recommendation quality
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The patent introduces asymmetry into the penalty function to differentiate between overrating and underrating content. By assigning different penalty weights to positive and negative prediction errors, the system can prioritize either avoiding uninteresting items (higher penalty for overestimations) or maximizing interesting items (higher penalty for underestimations). This asymmetric approach maintains overall prediction accuracy while improving user satisfaction by aligning the recommendation strategy with user preferences.

Inventive Principle:
Principle #4Asymmetry

Solution Approach 2:

The system dynamically adjusts the penalty function parameters based on user feedback and preferences. By making the penalty weights adaptable rather than fixed, the system can respond to changing user needs and optimize the balance between minimizing prediction error and maximizing user satisfaction. This dynamic adjustment resolves the contradiction by allowing the system to maintain accuracy while adapting to individual user requirements.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP2251994B1System and method for optimization of content recommendation
Publication Date: 2013.10.30 ADVANCED DIGITAL BROADCAST
  • EP2251994B1 patent drawingFigure 1
  • EP2251994B1 patent drawingFigure 2

AI summary

A method for optimization of a recommendation score of an item from a plurality of items, the method comprising the steps of receiving, from a recommender engine, a predicted recommendation score PS for the item and an uncertainty factor UF for the recommendation score, determining a correction coefficient CC, calculating an optimal correction OC by multiplying the uncertainty factor UF by the correction coefficient CC, correcting the predicted recommendation score PS by the optimal correction OC to obtain an optimized recommendation score OS, outputting the optimized recommendation score OS.