Attribute-Based Recommendation System Using Hierarchical Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data mining techniques for generating purchase recommendations lack granularity, leading to limited pattern correlation and overwhelming customers with numerous recommendations that do not adequately consider customer needs, especially as the complexity of products and services increases.
Innovation Solution
A system and method using an attribute-based recommendation domain model that normalizes customer, catalog, and transaction data to identify specific and actionable recommendations, prioritizing them using scoring criteria and generating targeted selling messages based on customer profiles and historical data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional data mining techniques use generic item descriptions (SKU level) to identify frequent sets, then the power-set/rule generation is tractable, but the frequent sets are not large and the quality of association rules is limited due to insufficient granularity
Solution Approach 1:
The patent segments items into multiple hierarchical levels (e.g., category, subcategory, brand, specific product) rather than treating them as single generic SKUs. This segmentation allows the system to work with coarser-grained categories for pattern discovery while still providing specific product recommendations, thus improving granularity without causing combinatorial explosion in rule generation.
Solution Approach 2:
The patent introduces a hierarchical dimension to item descriptions, organizing items from general categories to specific products. This dimensional change allows the system to discover patterns at the category level (reducing complexity) while still making specific recommendations (improving granularity) by traversing the hierarchy.
2Adaptability or versatility
If the number of services and products increases to meet customer needs, then the system can provide more comprehensive recommendations, but the complexity of making recommendations increases and the number of recommendations becomes overwhelming
Solution Approach 1:
The patent segments the recommendation process into distinct stages: pattern discovery at the category level, candidate generation, filtering based on customer profile, and final recommendation. This segmentation reduces the complexity of handling large numbers of products by breaking down the overwhelming task into manageable steps.
Solution Approach 2:
The patent extracts only the most relevant patterns and candidates from the large product set based on customer profiles and transaction history. By taking out only the necessary information for each customer rather than presenting all possible recommendations, the system maintains comprehensiveness while reducing complexity and avoiding overwhelming the customer.
3Reliability
If data mining techniques discover more association rules to cover complex products, then the coverage of recommendations improves, but the number of recommendations becomes overwhelming and customers are confused
Solution Approach 1:
The patent applies local quality by customizing recommendations based on individual customer profiles, purchase history, and preferences. Instead of applying the same set of association rules to all customers, the system tailors the presentation and selection of recommendations to each customer's specific context, improving clarity while maintaining comprehensive coverage.
Solution Approach 2:
The patent uses partial action by selecting and presenting only a subset of the discovered association rules that are most relevant to each customer. Rather than presenting all possible recommendations from the mined rules, the system filters and prioritizes them based on customer relevance, maintaining coverage while avoiding overwhelming the customer with too many options.
Data Source
AI summary
A data processing system generates recommendations for on-line shopping by scoring recommendations matching the customer's cart contents product page, configuration page, email and/or history using predetermined scoring criteria, such as item availability, item price, relevance of the item to the customer, item clarity, item exposure, total purchase amount in the cart, margin, recommendation confidence and lift, shopper profile matches, seasonality, customer history, time dependency, product marketing emphasis, substitutability, item popularity and/or product attributes associated with the recommended item. The scored recommendations may be ranked and/or filtered to prioritize the recommendations so that the best recommendation is selected for presentation to the customer. In addition, one or more scoring criteria or product attributes for the recommended item may be used to select a specific selling point/message text for inclusion with the selected recommendation.