Progressive ML Training for Auto Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models often rely on training data that may not accurately reflect user preferences, leading to inaccuracies in recommendations due to discrepancies between expressed preferences and actual behavior.
Innovation Solution
A computing device receives search data and historical vehicle financing data to generate training data, which is used to train a machine learning model. The model is further trained based on differences between actual vehicle purchases and expressed shopping preferences, adjusting weights in the neural network to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained using expressed user preferences (search data), then the model can provide recommendations based on user interests, but the accuracy of recommendations deteriorates when actual user behavior differs from expressed preferences
Solution Approach 1:
The patent implements a feedback mechanism where actual user purchase behavior is fed back into the training process. The system compares expressed preferences (search data) with actual behavior (purchase data) and uses this feedback to retrain the machine learning model, progressively improving recommendation accuracy by learning from real user actions rather than relying solely on stated preferences.
Solution Approach 2:
The patent changes the training parameters by incorporating multiple data sources with different weights. Instead of training solely on expressed preferences, the system adjusts training parameters to include actual purchase behavior data, changing the weight given to different data types during model training to better reflect real user behavior patterns.
2Quantity of substance
If training data is collected from user searches and browsing behavior, then the model can learn user interests, but errors in training data labeling cause inaccuracies in model output
Solution Approach 1:
The system uses actual purchase data as feedback to correct and refine training data. When users purchase vehicles different from what they searched for, this feedback signal helps the system identify and correct inaccuracies in the training data, progressively improving the quality of training data while maintaining large volume.
Solution Approach 2:
The patent creates a more accurate copy of user preferences by combining multiple data sources. Instead of relying on a single data type (search queries), the system creates an enhanced training data copy that incorporates both expressed preferences and actual behavior, resulting in a more accurate representation of true user interests.
3Device complexity
If the machine learning model is trained only on expressed shopping preferences, then the training process is simpler, but the model fails to capture actual user behavior patterns
Solution Approach 1:
The patent merges multiple data sources into a unified training approach. It combines expressed preferences (search data) with actual behavior (purchase data) into a single, comprehensive training process, creating a more reliable model that captures both stated and actual user interests while maintaining manageable complexity through integrated processing.
Solution Approach 2:
The system performs preliminary analysis of actual purchase behavior before final model training. By pre-processing and analyzing real user actions separately, the system can identify behavior patterns and use these as preliminary training signals, simplifying the overall training process while improving reliability.
Data Source
AI summary
Methods, systems, and apparatuses are described herein for progressively training an auto recommendation machine learning model based on differences between customer automobile searching preferences and actual automobile purchasing behavior. Search data indicating a history of auto shopping searches by one or more users may be received. Training data may be generated based on the search data and historical vehicle financing data and used to train a machine learning model. Auto shopping preference information may be received and provided as input to the trained machine learning model, which may output one or more recommended automobiles. After display of those one or more recommended automobiles, the system may receive an indication of an automobile purchased by a user and determine a difference between the automobile purchased by the user and the automobile shopping preference information. Based on that difference, the trained machine learning model may be further trained.