Client-Side Classification Model for Retail Item Set Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of large inventories and numerous possible item sets in retail facilities complicates the automated generation of recommended item sets for customers, leading to high computational costs and reduced likelihood of customer adoption.
Innovation Solution
A system that uses a classification model deployed on client devices to generate item set recommendations based on customer activity data, reducing the search space and increasing accuracy by employing historical data and specially configured item set definitions, and includes a server managing inventory and item set repositories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated generation of item set recommendations is implemented using traditional methods, then recommendations can be generated, but computational costs become prohibitively high
Solution Approach 1:
The patent segments the recommendation generation process into two phases: an offline phase where a classification model is trained using historical data and item set definitions, and an online phase where the trained model quickly generates recommendations for new item sets. This segmentation allows complex computations to be performed once during training, while inference during customer interactions remains computationally efficient.
Solution Approach 2:
The system performs preliminary training of the classification model using historical customer data and item set definitions before actual recommendation generation is needed. By pre-processing and training the model in advance, the system avoids the need for expensive real-time computations during customer interactions, significantly reducing operational computational costs.
2Productivity
If automated generation of item set recommendations is implemented, then recommendations can be generated, but accuracy is reduced due to complexity of large inventories
Solution Approach 1:
The classification model is trained using historical data that captures feedback patterns from past customer purchases and item set selections. By analyzing this feedback data, the model learns which item sets customers are most likely to purchase together, enabling accurate recommendations even for large inventories with numerous possible combinations.
Solution Approach 2:
The system transforms the recommendation problem from a combinatorial optimization problem to a classification problem by changing the parameter space. Instead of evaluating all possible item set combinations, the system classifies item sets into categories based on their association with customer purchase patterns, significantly improving accuracy while reducing computational complexity.
3Productivity
If traditional recommendation methods are used, then recommendations can be generated, but customer adoption is reduced
Solution Approach 1:
The system automatically generates and presents item set recommendations to customers without requiring manual intervention or complex user input. The classification model autonomously analyzes customer behavior patterns and provides personalized recommendations, making the system self-serving and improving customer adoption by eliminating the need for customers to manually search through large inventories.
Data Source
AI summary
A method includes: capturing, via a sensor of a mobile computing device, a first item identifier; determining, at the mobile computing device, whether the first item identifier matches one of a plurality of primary item identifiers; in response to determining that the first item identifier matches one of the primary item identifiers, providing input data to a classification model, the input data including the first item identifier; obtaining, via execution of the classification model, an item set identifier corresponding to an item set containing the first item identifier; and presenting the item set identifier at an output of the mobile computing device.