Mixture of Heterogeneous Experts for Fleet Vehicle Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current recommender systems for motor vehicle companies struggle to effectively recommend add-on vehicle services, particularly for fleet purchases, as they fail to consider the specific needs and preferences of bulk customers, such as vehicle type, region, and industry, leading to suboptimal selections for both customers and manufacturers.
Innovation Solution
A Mixture of Heterogeneous Experts (MoHeE) model-based system that automatically trains and selects multiple recommendation systems using historic user inputs to generate relevant recommendations by optimizing a weighted combination of metrics like precision, recall, and novelty, tailored to specific user preferences and vehicle-related services.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single recommendation system is used, then the system complexity is low, but the recommendation relevance and precision are insufficient for diverse fleet customer needs
Solution Approach 1:
The patent divides the recommendation system into multiple specialized models (MoHeE-FL, MoHeE-IND, MoHeE-REG) that each handle specific customer segments. Each model is trained on domain-specific data and optimized for particular fleet customer characteristics, allowing high precision recommendations without requiring a single overly complex universal system.
Solution Approach 2:
The system dynamically adjusts recommendation parameters based on customer attributes such as fleet size, industry sector, and geographic region. By changing the active model and weighting parameters according to customer profile, the system achieves high relevance while maintaining manageable complexity through parameterization rather than structural complexity.
2Measurement precision
If multiple specialized recommendation models are trained for different customer segments, then the recommendation precision improves, but the training time and computational resources increase
Solution Approach 1:
The patent performs preliminary training of multiple specialized models during the development phase, with each model pre-trained on specific domain data (fleet, industry, regional). This preliminary action allows the system to serve diverse customer segments with high precision without requiring time-consuming training during actual service delivery, as the models are ready to infer recommendations immediately.
Solution Approach 2:
The system creates specialized models by copying and adapting a base recommendation framework rather than developing entirely new systems from scratch. Each specialized model inherits the core architecture and can be trained independently on domain-specific data, reducing overall training time compared to building separate systems for each customer segment.
3Measurement precision
If the system considers multiple customer attributes (vehicle type, region, industry), then the recommendation accuracy improves, but the data processing complexity increases
Solution Approach 1:
The patent segments the data processing task by creating separate models for different customer attributes (MoHeE-FL for fleet, MoHeE-IND for industry, MoHeE-REG for region). Each model processes only the relevant attributes for its specific segment, reducing the complexity of data processing for any single model while maintaining high accuracy through specialized processing of relevant features.
Solution Approach 2:
The patent employs a universal MoHeE framework that can handle multiple customer attributes and segments through a common architectural structure. This multi-functional framework processes different attribute types (vehicle, regional, industry) using the same core mechanisms, reducing overall processing complexity compared to separate specialized processing systems.
Data Source
AI summary
The computing device obtains a training data set related to a plurality of historic user inputs associated with preferences of one or more services or items from an entity. For each of the one or more services or items, the computing device executes operations to train a plurality of models using the training data set to generate a plurality of recommended models, apply a validation data set to generate a plurality of predictions from the plurality of recommended models, obtain a weight of each metric of a plurality of metrics from the entity, obtain user inputs associated with user preferences, and determine a relevancy score for each metric. The computing device selects a recommended model based on the relevancy score of the selected metric or a combination of selected metrics, generates one or more recommendations for the users, and outputs the one or more generated recommendations to the users.