Multi-objective Recommendation System Using Segmented Predictive Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data analytics systems struggle to provide practical recommendations in real-life scenarios, as they often focus on single objectives, leading to impractical trade-offs between multiple conflicting goals, such as maximizing speed while maintaining safety and fuel efficiency.
Innovation Solution
The system allows users to specify multiple objectives and uses a multi-objective recommender component that determines candidate changes based on both greedy and evolutionary algorithms, generating recommendations that balance these objectives by predicting changes that improve the record with respect to multiple objectives simultaneously.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If single-objective optimization is used to maximize a specific field, then the optimization speed is fast and the solution is simple, but the recommendations become impractical due to ignoring trade-offs with other conflicting objectives
Solution Approach 1:
The patent segments the optimization process into two distinct phases: first, multiple single-objective predictive models are trained independently for each objective field; second, these models are integrated through a multi-objective scoring function that combines individual scores into a comprehensive evaluation. This segmentation allows fast single-objective optimization while achieving practical multi-objective balance.
Solution Approach 2:
The patent merges multiple single-objective predictive models into a unified multi-objective recommendation system. The scoring function combines predictions from individual models (e.g., speed model, safety model, fuel efficiency model) to generate a comprehensive score that reflects trade-offs across all objectives, producing practical recommendations that consider multiple conflicting goals simultaneously.
2Reliability
If multi-objective optimization is implemented to balance multiple conflicting goals, then the practicality of recommendations improves, but the system complexity increases
Solution Approach 1:
The system segments complexity management by training separate single-objective predictive models for each objective field independently. This modular approach allows each model to focus on one objective without being burdened by the complexity of balancing multiple conflicting objectives, while the final integration through scoring functions combines these simplified models into a comprehensive multi-objective system.
3Measurement precision
If comprehensive analysis of multiple objectives is performed, then the quality of recommendations improves, but the computational resources and time required increase
Solution Approach 1:
The patent applies preliminary action by pre-training multiple single-objective predictive models before the actual recommendation generation. These pre-trained models capture the relationships between actionable fields and each objective field independently. During recommendation generation, the system only needs to query these pre-trained models and combine their predictions through scoring functions, avoiding the need for complex real-time multi-objective optimization computations.
Data Source
AI summary
A method to provide multi-objective recommendations. The method includes receiving user input indicating a plurality of objectives, where each of the plurality of objectives indicates a desired goal for a field of interest, receiving user input indicating a plurality of actionable fields, receiving user input indicating selection of one of a plurality of records in a data set, determining, based on applying an evolutionary algorithm, one or more candidate changes to values of the plurality of actionable fields of the selected record, determining, for each of the one or more candidate changes, a multi-objective score for that candidate change, selecting one or more of the one or more candidate changes to recommend to a user based on the multi-objective scores of the one or more candidate changes, and providing, for display to the user, the selected one or more candidate changes as recommended changes.