Pairwise Comparison Data Mining for Item Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data mining methods for detecting behavioral relationships between items in electronic catalogs and other systems often fail to effectively help users discriminate between alternative items, relying solely on user descriptions and ratings, and can result in poor bundling suggestions.
Innovation Solution
Generating pairwise comparison data by analyzing user activity to determine the preference of users who consider multiple items, presenting this data to users to assist in making informed selection decisions, and identifying item pairs suitable for purchase or acquisition in combination by using both viewing and purchase histories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If data mining processes identify items commonly viewed or purchased together, then users are assisted in identifying candidate items, but users cannot effectively discriminate between these candidate items
Solution Approach 1:
The patent segments the item selection process into two distinct phases: (1) identifying candidate items using traditional data mining methods, and (2) discriminating between candidates using pairwise comparison data. This segmentation allows each phase to be optimized independently, with the first phase providing broad candidate identification and the second phase enabling focused discrimination through direct item comparisons.
Solution Approach 2:
The patent introduces pairwise comparison data as an intermediary element between item identification and final selection. This intermediary provides users with direct comparative information about specific item pairs, serving as a bridge that connects the broad set of candidate items to the final discrimination decision, thereby resolving the inability to effectively distinguish between candidates.
2Extent of automation
If purchase-based item relationships are used to suggest item pairs for bundling, then automated bundling suggestions are provided, but the suggestions are of poor quality
Solution Approach 1:
The patent changes the parameters used for generating bundling suggestions from simple purchase co-occurrence metrics to a more sophisticated analysis that incorporates pairwise comparison data and user preference information. By altering the underlying parameters from basic transaction data to preference-based comparison data, the system maintains automation while significantly improving suggestion quality.
Solution Approach 2:
The patent incorporates feedback loops where pairwise comparison results and user selection patterns are continuously analyzed to refine bundling suggestions. The system uses feedback from actual user comparisons and selections to improve the accuracy of future bundling recommendations, thereby enhancing quality while maintaining automated operation.
3Device complexity
If users rely solely on item descriptions, ratings, and reviews for selection, then no additional data processing is required, but users cannot make informed discrimination decisions
Solution Approach 1:
The patent performs preliminary data processing to generate pairwise comparison data before the user needs to make selection decisions. By pre-computing and organizing comparison information between item pairs, the system prepares discrimination-ready data in advance, allowing users to access refined comparison results without experiencing the complexity of raw data processing during their decision-making process.
Data Source
AI summary
Data mining systems and methods are disclosed for generating data that is helpful to users in selecting between items represented in an electronic data repository, such as an electronic catalog. One disclosed data mining method generates pairwise comparison data for particular pairs of items. The pairwise comparison data for a given item pair reveals the degree to which users who consider both items in the pair select one item over the other. The pairwise comparison data may be appropriately exposed to users of the electronic repository. For instance, in the context of an electronic catalog, an item detail page for item A may be supplemented with an indication that users who view both item A and item B select item B a specified percentage of the time. Another data mining method uses item viewing histories and item purchase histories of users in combination to identify pairs of items that are good candidates for being recommended in combination.