Hybrid Recommendation System Using OLAP and Multi-Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional recommendation systems based on Collaborative Filtering (CF), Information Filtering (IF), and Association Rule based Filtering (RF) face limitations in accuracy and scalability, including the early rater problem, sparsity, poor focus, and inefficiency in generating relevant recommendations, especially when lacking content information and cross-references.
Innovation Solution
The method employs On-Line Analytical Processing (OLAP) for multilevel and multidimensional data analysis, dimensioning similarity and association rule matrices to generate top-N recommended products within specific scopes, such as categories and customer groups, using multilevel and multidimensional CF, RF, and IF, and integrates neighborhood-oriented rule mining for personalized recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a pure Collaborative Filtering (CF) based recommendation system is used, then recommendations are generated based on customer similarity, but it suffers from the early rater problem, sparsity problem, and poor focus for items with too few ratings
Solution Approach 1:
The patent combines Collaborative Filtering (CF), Information Filtering (IF), and Association Rule based Filtering (RF) into a unified recommendation system. This integration allows the system to leverage customer similarity (CF), item content analysis (IF), and cross-selling patterns (RF) simultaneously, thereby overcoming the limitations of each individual approach and providing reliable recommendations even for items with limited ratings.
Solution Approach 2:
The recommendation system uses a composite approach by integrating multiple filtering techniques (CF, IF, RF) into a single hybrid system. This composite structure allows the system to benefit from the strengths of each filtering method while mitigating their individual weaknesses, particularly the sparsity and early rater problems associated with pure CF systems.
2Reliability
If a pure Information Filtering (IF) based recommendation system is used, then recommendations are generated based on item content profiles, but it is restricted to referring to items similar to those already rated (over specification problem)
Solution Approach 1:
The patent merges Information Filtering (IF) with Collaborative Filtering (CF) and Association Rule based Filtering (RF). This combination allows the system to use item content profiles (IF) while also incorporating customer behavior patterns (CF) and cross-selling associations (RF), thereby overcoming the over-specification limitation and enabling recommendations for a broader range of items beyond just content-similar ones.
Solution Approach 2:
The system adds dimensional diversity by integrating multiple filtering approaches that operate on different dimensions: CF operates on customer-customer similarity, IF on item-content similarity, and RF on item-item association. This multi-dimensional approach allows the system to break free from the single-dimension constraint of pure IF and provide versatile recommendations across different item types.
3Reliability
If a pure Association Rule based Filtering (RF) system is used, then recommendations are generated based on item correlation, but many rules may not be relevant to the given customer and selecting rules with fixed thresholds leads to too many or too few rules
Solution Approach 1:
The patent combines Association Rule based Filtering (RF) with Collaborative Filtering (CF) and Information Filtering (IF). This integration allows the system to use association rules (RF) while filtering and prioritizing them based on customer similarity (CF) and item content relevance (IF). This combination solves the relevance problem by ensuring that only customer-specific and contextually appropriate rules are applied.
Solution Approach 2:
The system applies local quality by customizing the application of association rules for each customer based on their profile and behavior patterns. Instead of applying fixed threshold rules uniformly to all customers, the system adapts rule selection and weighting to each customer's local context, thereby improving relevance and ease of operation.
4Ease of operation
If recommendation systems use fixed minimal support and confidence thresholds for association rules, then rule selection is simplified, but it leads to either too many or too few rules being generated
Solution Approach 1:
The patent implements dynamic threshold adjustment by integrating multiple filtering processes that adaptively determine rule relevance. Instead of using fixed minimal support and confidence thresholds, the system dynamically adjusts rule selection based on customer similarity (CF), item content analysis (IF), and contextual factors (RF), thereby generating an appropriate number of relevant rules while maintaining ease of operation.
Data Source
AI summary
A method and system for generating recommendations are described. The method of generating a recommendation list of products to a customer comprises using OLAP (On-Line Analytical Processing) to analyze raw data in a multilevel and multidimensional manner. Furthermore, the method includes applying a plurality of filtering processes to the raw data to generate a plurality of preliminary recommendation lists. Moreover, the method further comprises generating the recommendation list based on the plurality of preliminary recommendation lists.