Weakly-Supervised Product Compatibility Recommendation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing product recommendation techniques face challenges in providing accurate and comprehensive compatibility predictions due to the lack of manually curated data, heterogeneity of product information, and inefficiencies in rule discovery methods, leading to incorrect data propagation and limited interpretability.
Innovation Solution
The method involves generating weakly labeled instances using user behavior data, iteratively weighing data sets to focus on large error instances, and applying decision-tree based rule generation with pretrained language models to extract hidden information from unstructured product descriptions, forming a weighted ensemble compatibility model for improved product recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manually curated data is used for training compatibility models, then data quality and reliability are improved, but data quantity and coverage are limited
Solution Approach 1:
The system performs preliminary actions by using pretrained language models to extract structured attributes from unstructured product descriptions before training the compatibility model. This preprocessing step creates a foundation of structured data that can be combined with limited manually curated data, effectively expanding the training dataset while maintaining quality standards.
Solution Approach 2:
The system enables self-service by automatically generating structured product attributes using pretrained language models without requiring manual annotation for each product. The model self-trains on extracted attributes and user behavior data, reducing dependency on manually curated data while maintaining reliability through iterative refinement.
2Ease of operation
If rule discovery methods are used to extract compatibility rules, then interpretability is improved, but correctness and completeness of rules are reduced due to incorrect data propagation
Solution Approach 1:
The system implements feedback mechanisms by iteratively refining compatibility rules based on model predictions and user behavior data. The system evaluates rule performance, identifies incorrect predictions, and uses this feedback to adjust and improve rule correctness while maintaining interpretability through the rule-based approach.
Solution Approach 2:
The system replaces traditional mechanical rule discovery methods with a hybrid approach that combines rule-based systems with pretrained language models. This substitution allows the system to leverage the pattern recognition capabilities of neural networks while maintaining the interpretability of rule-based systems, reducing incorrect data propagation.
3Loss of information
If pretrained language models are used to extract attributes from unstructured product descriptions, then information extraction capability is improved, but processing time and computational resources increase
Solution Approach 1:
The system performs preliminary action by using pretrained language models during the offline training phase to extract structured attributes from unstructured product descriptions. This preprocessing creates a rich feature set that can be reused during online inference, reducing processing time when actual recommendations are generated while maximizing information extraction during the preparation phase.
Data Source
AI summary
A computer implemented method for determining object compatibility includes obtaining a first data set, a second data set, and a first compatibility model. The method also includes determining, by the compatibility system, error instances in the second data set by applying the first compatibility model to the second data set. The method further includes determining, by the compatibility system, labeling rules based on the error instances, and determining a third data set by applying the labeling rules to the first data set. The method also includes determining, by the compatibility system, a second compatibility model based on the third data set and determining an ensemble compatibility model based on the first compatibility model and the second compatibility model. The method further includes determining, by the compatibility system, a product recommendation based on the ensemble compatibility model and a user selection of a first product.