Social Network Recommendation Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation systems in social networks often overlook anomalous features of user selections, leading to missed recommendations that could be of interest due to being overwhelmed by non-anomalous features.
Innovation Solution
A method that determines and prioritizes features of a selected product, calculating correlation values with previously purchased items to identify anomalous features, and makes recommendations based on products or network connections sharing these features, giving higher priority to low correlation values indicative of significant anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If recommendation systems consider multiple features or characteristics of selected and unselected items together in calculating recommendation values, then the system can provide comprehensive recommendations, but non-anomalous features overwhelm anomalous features that may have great significance to the member
Solution Approach 1:
The patent extracts and isolates anomalous features from the overall feature set by calculating anomaly scores that measure how much each feature deviates from the member's historical pattern. This extraction allows the system to separately identify and prioritize significant anomalies that would otherwise be overwhelmed by numerous non-anomalous features in comprehensive recommendation calculations.
Solution Approach 2:
The patent applies local quality by assigning different weights to different features based on their anomaly scores. Instead of treating all features uniformly, the system identifies specific local features (those with high anomaly scores) and gives them elevated importance in the recommendation calculation, allowing significant anomalies to influence recommendations appropriately while maintaining consideration of other features.
2Stability of the object's composition
If recommendation systems prioritize items most similar to those already selected by the member, then the system maintains consistency with user preferences, but items with anomalous features that may be of great interest to the member are not included among recommendations
Solution Approach 1:
The patent introduces dynamics by making feature weights adaptive rather than static. The system continuously calculates anomaly scores based on the member's evolving selection history, allowing the importance of different features to change dynamically over time. This enables the system to adapt to new interests and anomalies while maintaining stability in the overall recommendation framework.
Solution Approach 2:
The patent changes parameters by transforming fixed similarity metrics into dynamic anomaly-based weights. Instead of using constant similarity thresholds, the system calculates anomaly scores that reflect the member's current preferences and uses these scores to adjust feature weights, thereby balancing consistency with the member's established patterns and adaptability to new, potentially significant interests.
3Ease of manufacture
If recommendation systems use traditional similarity-based approaches, then the implementation is straightforward and computationally efficient, but the system fails to identify and recommend items with significant anomalous features
Solution Approach 1:
The patent introduces an intermediary component - the anomaly score calculation mechanism - that bridges traditional similarity-based approaches and advanced anomaly detection. This intermediary layer processes feature data to generate anomaly scores, which then inform the recommendation ranking. This approach maintains the simplicity of traditional methods while adding the precision of anomaly detection through a modular intermediate step.
Data Source
AI summary
Embodiments of the invention provide methods and program products for making a recommendation to a purchaser and/or member of a social network. A first aspect of the invention provides a method of making a recommendation to a purchaser, the method comprising: determining a plurality of features of a first product selected by a purchaser; prioritizing the plurality of features of the first product; and making at least one recommendation to the purchaser, the at least one recommendation being selected from a group consisting of: a second product sharing at least one feature of the first product and a social network connection determined to have purchased another product sharing at least one feature of the first product.