Recommender System Catalog Clustering for Relevance

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The abundance of information available today overwhelms users with irrelevant data, making it challenging for recommender systems to effectively filter and recommend relevant items based on user preferences.

Innovation Solution

A recommender system that clusters items based on shared preferences and characteristics of a population, using explicit and implicit data to associate user legacy items with catalog clusters and recommend items from both tagged and related clusters, applying constraints such as inner product magnitudes and content-based filtering algorithms.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If recommender systems process abundant information to provide comprehensive recommendations, then the quantity of recommended items increases, but the relevance and quality of recommendations deteriorates due to information overload

Engineering Contradiction:

Improvequantity of recommended itemsVSAvoidrelevance of recommendations

Core Design Contradiction:

Quantity of substanceVSReliability

Solution Approach 1:

The patent segments the catalog of items into multiple clusters based on shared characteristics and user preferences. Each cluster represents a coherent group of items (e.g., movies by genre, products by category), allowing the system to manage abundant information in organized units rather than as an overwhelming whole, thus maintaining recommendation quality while increasing quantity

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system performs preliminary clustering of all catalog items into coherent groups before generating recommendations. This pre-organization of information into clusters based on item characteristics and user preferences enables efficient filtering and selection, allowing comprehensive recommendations to be generated without overwhelming the user or sacrificing relevance

Inventive Principle:

Principle #10Preliminary action

2Reliability

If the system filters information to improve recommendation relevance, then the quality of recommendations improves, but the quantity of available options decreases

Engineering Contradiction:

Improverelevance of recommendationsVSAvoidquantity of recommended items

Core Design Contradiction:

ReliabilityVSQuantity of substance

Solution Approach 1:

The system dynamically adjusts the number and composition of recommended items based on user preferences and cluster associations. By leveraging multiple tagged clusters related to user legacy items, the system can flexibly expand or contract the recommendation set to balance relevance and quantity according to individual user needs and context

Inventive Principle:

Principle #15Dynamics

3Measurement precision

If the system processes explicit and implicit user data to improve personalization, then the accuracy of recommendations improves, but the complexity of the system increases

Engineering Contradiction:

Improveaccuracy of user preference detectionVSAvoidsystem processing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces catalog clusters as intermediary structures between raw user data and final recommendations. By associating user legacy items with clusters and then selecting items from related clusters, the system mediates the complex processing of explicit and implicit user data through a structured intermediate representation, reducing overall system complexity while maintaining personalization accuracy

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS10438268B2Recommender system

Publication Date: 2019.10.08 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

Embodiments of the invention provide methods and apparatus for recommending items from a catalog of items to a user by parsing the catalog of items into a plurality of catalog clusters of related items and recommending catalog items to the user from catalog clusters to which items previously preferred by the user belong.