Cross-domain Recommender Systems Using Domain Separation Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional information recommendation technologies face challenges in accurately estimating user-relevant information due to insufficient selection history and cross-domain recommendation difficulties, especially when there are few users sharing services across domains.
Innovation Solution
A training apparatus and method that utilizes a model with input, intermediate, and output layers to learn feature information from one domain and corrects it based on feature information from another domain, enhancing estimation accuracy by integrating data from both domains through neural networks and autoencoders.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional co-occurrence learning is used to estimate user-relevant information, then the system can provide recommendations based on user selection history, but the accuracy decreases when the selection history is insufficient
Solution Approach 1:
The system performs preliminary learning in a first domain to extract feature information before applying it to the second domain. This preliminary action allows the system to build a knowledge base in advance that can be transferred to domains with insufficient data, thereby maintaining estimation accuracy even when selection history is limited.
Solution Approach 2:
Feature information extracted from the first domain serves as an intermediary that bridges the gap between domains with sufficient data and domains with insufficient data. This intermediary enables the system to transfer knowledge across domains, improving recommendation accuracy in the second domain without requiring extensive local selection history.
2Adaptability or versatility
If cross-domain recommendation is implemented with few users commonly using each service, then the system can provide cross-service recommendations, but the estimation accuracy becomes difficult to achieve
Solution Approach 1:
The patent uses feature information as an intermediary to transfer knowledge between domains with few common users. By extracting domain-specific features from the first domain and applying them to the second domain, the system maintains recommendation accuracy even when user overlap between services is minimal.
Solution Approach 2:
The system changes the parameter representation by extracting and transforming feature information between domains. This parameter transformation allows the system to adapt recommendations to different domains while maintaining accuracy, effectively handling the cold start problem in cross-domain scenarios.
3Measurement precision
If domain-specific feature learning is performed, then the system can capture domain-specific patterns, but the system complexity increases due to multiple domains and models
Solution Approach 1:
The patent segments the learning process into distinct domains with separate feature extraction mechanisms. By dividing the problem into first domain learning and second domain application, the system manages complexity through modular design while maintaining high domain-specific accuracy.
Data Source
AI summary
A training apparatus includes an acquiring unit that acquires a first model including an input layer to which input information is input; a plurality of intermediate layers that executes a calculation based on a feature of the input information that has been input; and an output layer that outputs output information that corresponds to output of the intermediate layer. The training apparatus includes a training unit that trains the first model such that, when predetermined input information is input to the first model, the first model outputs predetermined output information that corresponds to the predetermined input information and intermediate information output from a predetermined intermediate layer among the intermediate layers becomes close to feature information that corresponds to a feature of correspondence information that corresponds to the predetermined input information.