Wide-and-Deep Similar Item Recommendation for Ad Relevance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Advertisement recommendation systems often provide irrelevant item advertisements, leading to lost sales and customer attrition, as customers may not purchase advertised items and may leave the website to find more relevant products elsewhere.
Innovation Solution
A computing device generates feature data based on historical interaction and transaction data, trains machine learning models, and ranks recommended items using a wide-and-deep-architecture to predict future purchase likelihood, filtering based on categorical attribute embeddings to ensure relevance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional advertisement recommendation systems are used, then advertisements can be displayed to customers, but the advertisements are often irrelevant to customers leading to lost sales
Solution Approach 1:
The system performs preliminary actions by training machine learning models in advance using historical interaction data, transaction data, and item categorical data. These pre-trained models are then executed to generate prediction data about future purchase likelihood, enabling the system to proactively identify relevant advertisements before they are displayed to customers, thereby improving both relevance and conversion rates
Solution Approach 2:
The system implements feedback mechanisms by using historical website interaction data and historical transaction data to continuously train and improve the machine learning models. The models learn from past customer behaviors and purchase patterns, adjusting their predictions to better match actual customer preferences, which enhances advertisement relevance over time and improves sales conversion
2Loss of time
If traditional advertisement recommendation systems are used, then advertisements can be displayed, but customers spend time reviewing irrelevant advertisements
Solution Approach 1:
The system replaces manual customer evaluation of advertisement relevance with an automated machine learning-based recommendation system. The trained models automatically analyze customer data and item attributes to generate ranked lists of recommended items, substituting the mechanical process of customer browsing and filtering with an intelligent automated system that delivers relevant advertisements upfront, reducing time waste and improving purchasing efficiency
Data Source
AI summary
A system including one or more processors and one or more non-transitory computer-readable media storing computing instructions that, when executed on the one or more processors, cause the one or more processors to perform functions comprising: receiving one or more vectors representing one or more types of features for a pair of items; generating, using a similarity item model of a machine learning architecture, a prediction for a similar item, wherein the similarity item model combines a pair of separately trained machine learning models; combining a first output of the gradient boosted model and a second output of the neural network model to generate a similarity score for the pair of items; and transmitting the similar item to a first position on a carousel display of a website that concurrently displays the anchor item on the website. Other embodiments are disclosed.