Bi-directional Collaborative Filtering for Online Network Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing collaborative filtering models in online networks operate with a one-way 'interest' relationship, which is insufficient for domains like product offerings where relationships are bi-directional, leading to ineffective recommendations.
Innovation Solution
Implementing a bi-directional collaborative filtering model that considers multiple directions of interaction, such as message exchanges and responses, to identify relevant information for users expanding their activity channels by aggregating attributes of target users who have engaged with similar users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a one-way collaborative filtering model is used, then the system is simple to implement, but the recommendation accuracy is insufficient for bi-directional relationships
Solution Approach 1:
The patent inverts the traditional one-way collaborative filtering approach by implementing bi-directional filtering that considers both outgoing and incoming interactions. Instead of only analyzing what users send or what users receive, the system simultaneously processes both directions of interaction to generate recommendations, thereby improving accuracy while maintaining reasonable system complexity
Solution Approach 2:
The patent adds a new dimension to the collaborative filtering model by incorporating the directionality of interactions. The system evaluates relationships from multiple directional perspectives (outgoing messages, incoming messages, responses) rather than treating all interactions as uniform, thereby capturing more nuanced relationship patterns for better recommendations
2Measurement precision
If bi-directional collaborative filtering is implemented, then recommendation accuracy improves, but system complexity increases
Solution Approach 1:
The patent segments the bi-directional filtering process into distinct components: identifying similar users based on outgoing interactions, identifying target users based on incoming interactions, and combining these results to generate recommendations. This segmentation allows the complex bi-directional analysis to be implemented through manageable, modular steps that reduce overall system complexity
Solution Approach 2:
The patent introduces intermediary data structures and processing layers that mediate between the raw interaction data and the final recommendations. These intermediaries organize and structure the bi-directional interaction data, making it easier to process and reducing the computational complexity of the overall system
3Productivity
If only one-way interest relationships are considered, then the model is computationally efficient, but it fails to capture bi-directional relationships in domains like product offerings
Solution Approach 1:
The patent performs preliminary actions by pre-identifying similar users and target users separately before generating recommendations. This preliminary segmentation allows the system to efficiently organize bi-directional relationship data in advance, reducing the computational burden during the actual recommendation generation process while preserving complete relationship information
Data Source
AI summary
In one technique, one or more search criteria associated with a first user are determined. Based on the one or more search criteria, a search of a first set of users is performed. As a result of the search, a subset of the first set of users is identified. A second set of users that have responded to messages from users in the subset is identified. Instances of attribute values of an attribute of the second set of users are aggregated to generate aggregated values, each of which corresponds to a different attribute value of the attribute values. Based on one or more of the aggregated values, content that is related to one or more attribute values that correspond to the one or more aggregated values is determined and presented to the first user.