Sequence Recommendation Modeling Complex Multi-Mode User Interests
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sequence recommendation methods fail to accurately model user interests due to their reliance on distinguishing between long-term and short-term interests, which do not effectively capture the dynamic and evolutionary nature of user preferences over varying sequence lengths.
Innovation Solution
A sequence recommendation method that extracts and models complex multi-mode user interests by separating historical item interaction sequences into long-term and short-term sequences, using multi-head self-attention modules to update embeddings, and fusing dynamic and static interests to capture evolutionary changes, resulting in more accurate personalized recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing methods distinguish between long-term and short-term interests using different sequence lengths, then user interests can be categorized, but the dynamic and evolutionary nature of user preferences cannot be accurately captured
Solution Approach 1:
The patent transforms the static distinction between long-term and short-term interests into a dynamic model by introducing evolutionary interests that capture how user preferences change over time. The interest evolution module dynamically adjusts interest representations based on temporal patterns, allowing the system to adapt to changing user preferences while maintaining accurate modeling of both stable and evolving interests.
Solution Approach 2:
The patent segments user interests into three distinct components: static interests (stable preferences), dynamic interests (temporally varying preferences), and evolutionary interests (changing patterns over time). This segmentation allows each component to be modeled with appropriate methods, resolving the contradiction by capturing both the categorical structure and dynamic nature of user preferences simultaneously.
2Adaptability or versatility
If sequence length is varied to capture different time horizons, then both long-term and short-term interests can be mined, but the representation accuracy of user interests deteriorates due to interest changes with length
Solution Approach 1:
The patent changes the parameter of interest representation from fixed-length sequences to evolving interest vectors that adapt to different temporal horizons. By modeling interests as evolving entities with explicit temporal dependencies, the system maintains accurate representations across varying sequence lengths without the degradation that occurs in traditional fixed-length approaches.
Solution Approach 2:
The patent introduces evolutionary interests as an intermediary layer between static and dynamic interests. This intermediary captures the transition patterns of user preferences over time, allowing the system to accurately represent user interests across different temporal scales by mediating between stable long-term preferences and varying short-term behaviors.
3Ease of manufacture
If traditional long-term and short-term interest modeling is used, then user preferences can be categorized, but the complex multi-mode nature of user interests cannot be accurately represented
Solution Approach 1:
The patent segments complex user interests into three manageable components: static interests, dynamic interests, and evolutionary interests. This segmentation makes the complex multi-mode nature of user interests tractable by breaking it down into distinct, modelable elements while maintaining high representation accuracy through the comprehensive coverage of different interest modes.
Data Source
AI summary
A sequence recommendation method based on extraction and modeling of complex multi-mode user interests is provided, including: obtaining long-term and short-term embedding sequences; obtaining updated long-term and short-term embedding sequences through the long-term and short-term embedding sequences; with embedding vectors of last items in the updated long-term and short-term embedding sequences as long-term and short-term dynamic interests of a user, obtaining long-term and short-term static interests of the user through weighted calculation; concatenating the long-term and short-term dynamic interests and the long-term and short-term static interests, and performing nonlinear change to obtain long-term and short-term evolutionary interests of the user; obtaining a dynamic interest, a static interest and an evolutionary interest of the user through element-wise summation; performing weighted summation to obtain a fused user interest; calculating a product of the fused interest with embedding of each item as a recommendation score of each item.