Cross-domain sequential recommendation method based on time series and projection enhancement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cross-domain recommendation systems face issues of negative transfer and failure to capture cyclical user preferences due to interference from unique information in the source domain, leading to inaccurate recommendations.
Innovation Solution
A cross-domain sequential recommendation method based on time series and projection enhancement, which encodes item nodes and timestamps separately, captures dependencies through directed matrices, aggregates information using multi-head attention, and employs a masking mechanism with contrastive learning to filter useless information and capture cyclical preferences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If source domain information is migrated to target domain, then data sparsity is alleviated, but negative transfer occurs due to unique source domain information interfering with target domain recommendations
Solution Approach 1:
The patent segments source domain information into shared information and unique information through domain-specific encoding layers. The shared information is migrated to the target domain while unique information is retained separately, preventing negative transfer from source domain specifics interfering with target domain recommendations.
Solution Approach 2:
The patent introduces a domain adapter as an intermediary component that selectively transfers information from source domain to target domain. This adapter uses attention mechanisms to identify and transfer only relevant shared information while filtering out domain-specific unique information that would cause negative transfer.
2Quantity of substance
If all source domain information is used, then information richness is maximized, but useless information increases causing negative transfer
Solution Approach 1:
The patent extracts and separates shared information from unique information in the source domain through domain-specific encoding and attention mechanisms. Only the extracted shared information is migrated to the target domain, while unique information is excluded to prevent negative transfer.
Solution Approach 2:
The patent applies different quality standards to different portions of source domain information. Shared information that is relevant across domains is migrated with high quality, while unique domain-specific information is either retained locally or discarded based on its relevance, ensuring only high-quality useful information affects target domain recommendations.
3Loss of information
If traditional cross-domain recommendation is used, then data migration is achieved, but cyclical user preferences are not captured
Solution Approach 1:
The patent incorporates time-series analysis with sinusoidal and cosine positional encodings to model periodic user preferences. This allows the system to capture cyclical patterns in user behavior such as seasonal preferences or recurring purchase cycles, improving preference capture accuracy beyond traditional migration methods.
Solution Approach 2:
The patent transforms static information migration into a dynamic process by incorporating temporal dimensions. The model adapts to changing user preferences over time through recurrent neural networks and time-aware attention mechanisms, allowing preferences to evolve rather than remain fixed during the migration process.
Data Source
AI summary
A cross-domain sequential recommendation method based on time series and projection enhancement. According to the present disclosure, first, single-domain and cross-domain interaction sequences are encoded by using a graph attention mechanism. To account for periodic variation of user preference, timestamp encoding is incorporated to capture temporal characteristics of user behavior. After combining the interaction sequences with the time encoding, a projection mechanism-based module is designed in the present disclosure to accurately capture unique features of a user in a specific domain and shared features in a mixed sequence, such that redundant information in a source domain is effectively prevented from being transmitted to a target domain. Finally, a contrastive learning auxiliary framework is designed to further enhance cross-domain sequence representation. The present disclosure effectively extracts the feature quality in each domain, mitigates the negative transfer through effective feature extraction, and ultimately improves the performance of cross-domain recommendation.