Self-Boosted Collaborative Filtering for B2B Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing collaborative filtering systems face challenges in providing accurate recommendations for business-to-business (B2B) transactions due to equal weightage given to historic purchases and neighbors, leading to sub-optimal product lists and neighbor sets, which increase computational burden and subjectivity in B2B scenarios with skewed purchase patterns.
Innovation Solution
The system employs self-boosted and principal component adjusted weights using an optimal neighbor set to generate recommendations, boosting historic weights and normalizing interactions via a principal component approach, considering an open sales pipeline for the immediate next quarter to improve collaborative filtering operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If equal weightage is given to historic purchases and neighbors in collaborative filtering, then the system is simple to implement, but recommendation accuracy deteriorates in B2B transactions with skewed purchase patterns
Solution Approach 1:
The patent applies local quality by differentiating the weightage given to historic purchases versus neighbor interactions. Specifically, historic purchases are assigned higher weightage (self-boosting) while neighbor interactions receive normalized weightage through principal component analysis. This localized differentiation resolves the contradiction by maintaining implementation simplicity while improving recommendation accuracy for B2B transactions with skewed purchase patterns.
2Measurement precision
If historic weights are boosted in B2B transactions, then recommendation accuracy improves, but computational burden increases
Solution Approach 1:
The patent applies parameter changes by transforming the weight parameters through principal component analysis. The historic weights are boosted using a self-boosting mechanism, and then the interaction matrix is normalized by applying principal component loadings as parameters. This parameter transformation reduces computational complexity while maintaining improved recommendation accuracy, as the complex weighted interactions are projected onto fewer principal components.
3Adaptability or versatility
If principal component analysis is applied to normalize interactions, then recommendation personalization improves, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-computing the principal component loadings from the interaction matrix before generating recommendations. The PCA is performed once to obtain the loading parameters, which are then reused for normalizing neighbor interactions during recommendation generation. This preliminary computation reduces the complexity of repeated PCA operations while maintaining high recommendation personalization for B2B transactions.
4Productivity
If an optimal neighbor set is used with self-boosted weights, then revenue coverage increases by 10%, but data processing requirements increase
Solution Approach 1:
The patent applies the extraction principle by isolating and processing only the most relevant data components. Specifically, the system extracts and processes historic purchase data with self-boosting, and separately processes neighbor interaction data through principal component analysis. By extracting and processing these data streams separately and selectively, the system achieves 10% higher revenue coverage while reducing overall data processing requirements compared to processing all data uniformly.
Data Source
AI summary
A system, method, and computer-readable medium are disclosed for performing a recommendation operation, comprising: optimizing a product list to provide an optimized product list for use when generating a recommendation for an account; optimizing a neighbor set to provide an optimized neighbor set for use when generating the recommendation for the account; boosting a self-cosine similarity metric to provide a boosted self-cosine similarity metric, the self-cosine similarity metric corresponding to the account; and, providing a recommendation for the account, the recommendation being based on the optimized product list, the optimized neighbor set and the boosted self-cosine similarity metric.