Cross-Media Recommendation Framework Using Shared Item Dictionaries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Newly launched social media sites face challenges in friend and item recommendations due to data sparsity and the lack of direct bridges between platforms, making it difficult to transfer knowledge effectively, as existing methods rely on anchor links and consistent item attributes which are often unavailable or inconsistent.
Innovation Solution
A cross-media joint friend and item recommendation framework that uses sparse transfer learning and cross-site rating and friend transfer learning to integrate within-platform correlations and cross-platform information, leveraging shared dictionaries and latent feature representations to bridge knowledge gaps between platforms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If knowledge transfer from mature sites is attempted using existing approaches, then recommendation performance can be improved, but the method requires overlapped users or similar items which are often unavailable or costly to identify
Solution Approach 1:
The patent introduces a shared item dictionary as an intermediary between different social media sites. This dictionary contains items that appear across multiple sites with consistent attributes, serving as a mediator to transfer knowledge without requiring direct user overlap. The shared dictionary enables indirect knowledge transfer by matching items across platforms rather than requiring direct user correspondence.
Solution Approach 2:
The patent extracts and isolates the item matching problem from the broader knowledge transfer challenge. By separating item attribute consistency verification from user overlap requirements, the system focuses on extracting reliable item information across sites. This extraction approach identifies a subset of items with consistent attributes that can serve as reliable bridges for knowledge transfer.
2Adaptability or versatility
If different schemes to show item attributes across sites are used, then site-specific customization is achieved, but attribute values become inconsistent, incomplete, and noisy
Solution Approach 1:
The patent applies local quality by treating different sites differently in terms of attribute handling. Each site maintains its own attribute schema and customization, but for items in the shared dictionary, the system enforces consistent attribute values. This allows site-specific flexibility while ensuring precision for cross-site matching items.
Solution Approach 2:
The patent changes the parameter representation by introducing a canonical attribute schema for items in the shared dictionary. While sites can use different attribute schemes locally, the system transforms attributes to a standardized format for cross-site comparison and knowledge transfer, ensuring consistency without losing site-specific customization capabilities.
3Measurement precision
If user-user relationships are exploited for item recommendation, then recommendation accuracy improves, but the ability to exploit user-item interactions for friend recommendation remains limited
Solution Approach 1:
The patent makes the user relationship model universal by applying the same latent factor decomposition technique to both user-user and user-item relationships. The framework treats friend recommendation and item recommendation symmetrically, using consistent mathematical formulations for both tasks. This enables bidirectional recommendation capability where user-item interactions can inform friend recommendations and vice versa.
Data Source
AI summary
Various embodiments of systems and methods for cross media joint friend and item recommendations are disclosed herein.