Interest Alignment Model for Cross-Domain Recommendation Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Personalized recommender systems face challenges in making accurate resource recommendations for objects with limited interaction data, as the resource recommendation model trained on abundant interaction data is not suitable for objects with scarce interaction data.
Innovation Solution
The proposed solution involves training an interest alignment model using an interest relationship between objects, which involves obtaining a dataset with both rich and scarce interaction data, performing feature extraction, and aligning interests across domains to reduce losses associated with feature extraction and interest alignment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a resource recommendation model is trained based on a large amount of interaction data of objects, then the recommendation effect is improved for objects with rich data, but the model performance deteriorates for objects with less interaction data
Solution Approach 1:
The patent introduces an interest alignment model as an intermediary that learns from multiple domains and aligns interest representations. This mediator enables knowledge transfer from domains with rich interaction data to domains with scarce data, resolving the contradiction between model accuracy on rich data and adaptability to scarce data scenarios
Solution Approach 2:
The patent creates a universal interest alignment model that can handle multiple domains simultaneously. The model learns common interest representations across different domains, making it versatile enough to serve both objects with rich interaction data and those with scarce data, thus achieving multi-functionality
2Reliability
If personalized recommendations are made using a resource recommendation model trained on abundant interaction data, then recommendation quality is improved for well-documented objects, but recommendation reliability deteriorates for objects with limited interaction data
Solution Approach 1:
The patent merges information from multiple domains by training the interest alignment model on cross-domain data. This combining approach enriches the information available for objects with limited interaction data by incorporating patterns and interests from other domains, thereby reducing data sparsity and improving recommendation reliability
3Measurement precision
If feature extraction is performed on domain data using an interest alignment model, then feature representation quality is improved, but computational complexity increases
Solution Approach 1:
The patent employs parameter sharing and alignment across domains in the interest alignment model. By changing the parameter structure to be shared and aligned rather than completely independent per domain, the model achieves better feature representation while controlling computational complexity through parameter efficiency
Data Source
AI summary
A method of model training includes obtaining a data set for training an interest alignment model of cross domain recommendation, the data set includes first domain data and second domain data. The method also includes performing a first feature extraction on the first domain data according to the interest alignment model to obtain a first domain feature representation and performing a second feature extraction on the second domain data according to the interest alignment model to obtain a second domain feature representation. Further, the method includes performing an interest alignment between the first domain and the second domain according to the interest alignment model based on the first domain feature representation and the second domain feature representation; and training the interest alignment model in a direction that reduces at least one of a first loss and a second loss, to obtain a trained interest alignment model.