Relationship Server Cross-Site Recommendation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Web sites face limitations in tracking end-user interactions across multiple platforms, leading to a restricted pool of relationship data for collaborative filtering, resulting in lower-quality recommendations due to the heterogeneity of data types and lack of comprehensive user profiling.
Innovation Solution
A relationship server is employed to collect and canonicalize end-user interactions from multiple web sites, creating unique identifiers for users and establishing relationships between arbitrary items, enabling collaborative filtering across diverse data types to generate high-quality recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a web site bases recommendations on only a limited set of relationship data from its own site, then the implementation is simple and device complexity is low, but the pool of relationship data is smaller and recommendation quality is lower
Solution Approach 1:
The patent introduces a relationship server as an intermediary component that collects, stores, and manages relationship data from multiple web sites. This mediator enables sites to access broader relationship data without directly implementing complex cross-site tracking infrastructure themselves, thus improving recommendation quality while keeping individual site complexity manageable
Solution Approach 2:
The relationship server provides universal functionality by serving multiple web sites with a shared relationship data pool. Instead of each site implementing its own isolated recommendation system, the universal relationship server handles data collection, storage, and recommendation generation for multiple sites, improving overall system efficiency and data utilization
2Quantity of substance
If a web site tracks user interactions across multiple web sites, then the pool of relationship data is larger and recommendation quality improves, but there is no convenient way to track interactions and implementation becomes more complex
Solution Approach 1:
The relationship server acts as a centralized intermediary that simplifies cross-site tracking by providing a single point of data collection and management. Sites interact with this mediator rather than implementing complex peer-to-peer tracking mechanisms, enabling broader data collection while managing implementation complexity
Solution Approach 2:
The system enables sites to automatically contribute their observed relationship data to the shared pool through standardized interfaces. Each site self-services by sending observed relationships to the relationship server, which automatically integrates them into the global dataset without requiring manual intervention or complex coordination
3Adaptability or versatility
If relationship data is limited to homogeneous types (e.g., only news stories or only consumer goods), then the data structure is simpler and easier to manage, but recommendations are homogenous and may be of less value
Solution Approach 1:
The relationship server implements parameter changes by introducing a typed item structure that can accommodate heterogeneous data types (news stories, consumer goods, newsgroups, images, etc.). By defining a flexible schema with type parameters, the system manages data heterogeneity systematically while enabling diverse recommendation possibilities across different item categories
Data Source
AI summary
A relationship server tracks end-user interactions across multiple web sites and generates recommendations. The web sites observe relationships established by end-user interactions. If end-users provide the same personally identifiable information to multiple web sites, the sites generate the same unique identifier for those end-users. The web sites send messages to the relationship server that reference the end-users using the identifiers and describe the relationships observed for the end-users. The relationship server receives messages from multiple web sites and canonicalizes them to produce an efficient representation of the relationships. Upon receiving a message requesting a recommendation based on an item, the relationship server performs collaborative filtering using the relationship data to identify a list of items to recommend. The relationship server sends the recommendations to the requesting entity and the recommendations are presented to the end-user.