Item Recommendation via Attribute Relevance Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation technologies, such as those using collaborative filtering, struggle to recommend diverse items when there is limited usage information, often resulting in users losing interest due to the presentation of only similar items.
Innovation Solution
An information processing apparatus and method that calculates an attribute relevance degree between items and attributes based on usage information and attribute correspondence information, selecting relevant attributes and forming candidate item aggregates to recommend items that are relevant to a reference item, even with limited usage data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If recommendation is based on keywords coincidence, then item relevance is improved, but item diversity deteriorates
Solution Approach 1:
The patent segments the recommendation process into two distinct stages: first identifying candidate items through attribute relevance calculation, then selecting final recommendations through keyword coincidence matching. This segmentation allows the system to prioritize attribute relevance for diversity while using keyword matching for precision, resolving the contradiction between relevance and diversity.
Solution Approach 2:
The patent applies different evaluation criteria to different stages of recommendation. Attribute relevance is used locally for candidate generation to ensure diversity, while keyword coincidence is applied locally for final selection to ensure precision. This local quality differentiation allows simultaneous optimization of both diversity and relevance.
2Measurement precision
If usage information is accumulated for collaborative filtering, then recommendation accuracy is improved, but system complexity deteriorates
Solution Approach 1:
The patent extracts and stores only the essential attribute correspondence information between items and attributes, rather than accumulating complete usage histories. This extraction approach maintains recommendation accuracy by preserving key relational data while significantly reducing system complexity and data storage requirements.
Solution Approach 2:
The patent creates a simplified representation of usage information through attribute correspondence tables that capture the essential relationships between items and attributes. This copying approach preserves the necessary information for accurate recommendations without storing the full complexity of usage histories.
Data Source
AI summary
A usage information storage section is configured to store usage information. An item information storage section is configured to store attribute correspondence information. A relevant attribute selection section is configured to calculate an attribute relevant degree based on the usage information and the attribute correspondence information, select a relevant attribute for each item identifier based on the calculated attribute relevant degree, and create relevant attribute information in which each item identifier and the selected relevant attribute are associated with each other. A relevant item selection section is configured to acquire attribute correspondence information including any attribute identifier included in the relevant attribute information, calculate a relevant degree between each item identifier included in a relevant candidate item aggregate as an aggregate of item identifiers included in the acquired attribute correspondence information and a reference item, and select an item relevant to the reference item based on the calculated relevant degree.