Unbiased Conversion Rate Item Recommendation System

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional e-commerce recommendation systems exploit the contrast effect to increase user interaction with recommended items, often at the expense of accuracy, as they focus on maximizing conversion rates without considering the unbiased conversion rates of items, leading to potential user exploitation.

Innovation Solution

A machine learning-based item recommendation system that selects a contrast group by determining an unbiased conversion rate for each item and minimizing the contrast effect, using a recommendation model that includes an item selection model and a contrast group selection model to recommend items with the highest unbiased conversion rate while controlling contrast bias.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional recommendation systems exploit the contrast effect by presenting recommended items with similar high-value and low-value contrast items, then the conversion rate of recommended items increases, but the recommendation accuracy and user fairness deteriorate

Engineering Contradiction:

Improveconversion rateVSAvoidrecommendation accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary estimation of unbiased conversion rates for all recommendable items before selecting the recommended item and contrast items. This preliminary action ensures that the item with the highest intrinsic conversion potential is selected as the recommended item, preventing exploitation of items with low genuine value merely because they benefit from contrast effects

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The recommendation system segments the conversion rate into two components: unbiased conversion rate (intrinsic item quality) and contrast effect (influence from surrounding items). By separating these components, the system can select the recommended item based on unbiased conversion rate while still utilizing contrast items to enhance overall conversion, thus resolving the contradiction between conversion rate and recommendation accuracy

Inventive Principle:

Principle #1Segmentation

2Measurement precision

If conventional recommendation systems present recommended items alone without contrast items, then recommendation accuracy is maintained, but the conversion rate and user interaction probability decrease

Engineering Contradiction:

Improverecommendation accuracyVSAvoidconversion rate

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The system applies different selection criteria to different items within the recommendation presentation: the recommended item is selected based on unbiased conversion rate (local quality of intrinsic value), while contrast items are selected based on their ability to create effective contrast (local quality of contextual influence). This differentiated approach maintains recommendation accuracy for the primary recommendation while boosting overall conversion through strategically selected contrast items

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

Contrast items serve as intermediaries that mediate between the recommended item and the user's decision-making process. They provide contextual reference points that help users evaluate the recommended item's value, thereby increasing conversion probability without compromising the accuracy of the primary recommendation

Inventive Principle:

Principle #24Intermediary (Mediator)

3Productivity

If conventional systems maximize conversion rates through contrast effect exploitation, then productivity increases, but user exploitation and ethical concerns arise

Engineering Contradiction:

Improveconversion rateVSAvoiduser exploitation

Core Design Contradiction:

ProductivityVSObject-affected harmful factors

Solution Approach 1:

The system performs a preliminary assessment of each item's unbiased conversion rate to identify items with genuine intrinsic value before presenting them as recommendations. This preliminary anti-action prevents the selection and promotion of items that would otherwise be chosen merely due to their ability to benefit from contrast effects, thereby preventing user exploitation while maintaining high conversion rates

Inventive Principle:

Principle #9Preliminary anti-action

Data Source

PatentUS20230142768A1Preventing contrast effect exploitation in item recommendations

Publication Date: 2023.05.11 ADOBE INC

AI summary

An item recommendation system receives a set of recommendable items and a request to select, from the set of recommendable items, a contrast group. The item recommendation system selects a contrast group from the set of recommendable items by applying a recommendation model to the set of recommendable items. The recommendation model includes an item selection model configured to determine an unbiased conversion rate for each item of the set of recommendable items and select a recommended item from the set of recommendable items having a greatest unbiased conversion rate. The recommendation model includes a contrast group selection model configured to select, for the recommended item, a contrast group comprising the recommended item and one or more contrast items. The item recommendation system transmits the contrast group responsive to the request.