Ad Generation System Using ML for Relevance and Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing online advertising methods struggle to effectively target users with ads that match their interests beyond contextual matching, limiting the number of ads businesses can place and relying heavily on manual intervention for ad creation and keyword association.
Innovation Solution
A system and method that uses machine learning algorithms to select and generate ads by leveraging existing product information, including taxonomy, prices, user reviews, and expert reviews, and incorporates endorsements from users and experts to enhance ad relevance and visibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning algorithms are used to dynamically create ads from product catalogs, then the quantity and relevance of ads increase, but the system complexity increases
Solution Approach 1:
The system enables automatic ad generation by having the machine learning model self-service the entire ad creation process. The model takes product catalog data as input and autonomously generates ad content, selects images, and determines pricing strategies without requiring manual ad creation for each product. This self-service approach resolves the contradiction by automating the complex tasks that would otherwise require significant human intervention.
Solution Approach 2:
The patent introduces an intermediary layer between the product catalog and the ad delivery system. This intermediary consists of the machine learning model that processes raw product data, extracts relevant features, and transforms them into ad-ready content. The intermediary handles the complexity internally, presenting a simplified interface to both the product catalog source and the ad delivery system, thus resolving the complexity issue while maintaining high productivity.
2Manufacturing precision
If manual intervention is used for ad creation and keyword association, then ad quality can be controlled, but the time and resources required increase significantly
Solution Approach 1:
The patent replaces the mechanical system of manual ad creation with an intelligent automated system. Instead of human advertisers manually writing ad copy, selecting images, and associating keywords, the machine learning model performs these tasks automatically. The model analyzes product catalogs, generates relevant ad content, and associates appropriate keywords based on learned patterns from training data, thereby maintaining quality control while dramatically reducing the time and resources required.
Solution Approach 2:
The system performs preliminary actions by pre-processing product catalog data and pre-training the machine learning model on extensive product information before actual ad generation is needed. The model is trained in advance on large datasets of product catalogs, ad performance data, and user behavior patterns. This preliminary training enables the model to quickly generate high-quality ads in real-time without requiring manual intervention during the actual ad creation process.
3Reliability
If contextual matching is used for ad targeting, then ads can be relevant to website content, but the ability to target user interests beyond context is limited
Solution Approach 1:
The patent adds another dimension to ad targeting by incorporating user profile data and historical behavior information alongside contextual matching. Instead of relying solely on the current website content, the system multi-dimensionally matches ads by considering: (1) contextual relevance to the current page, (2) user's historical ad interactions and click patterns, (3) user's demographic and preference profiles, and (4) product catalog attributes. This multi-dimensional approach simultaneously improves ad relevance and expands targeting capability beyond what contextual matching alone can achieve.
Solution Approach 2:
The machine learning model serves multiple functions simultaneously: it acts as a contextual analyzer, a user behavior predictor, a product matcher, and an ad optimizer. This multi-functional model can process various types of input data (product catalogs, user profiles, contextual information) and generate optimized ad content that satisfies multiple targeting criteria at once. The universality of the model enables it to adapt to different targeting scenarios while maintaining reliable ad relevance across diverse contexts.
Data Source
AI summary
Embodiments of a system method and computer program product for selecting an advertisement and presenting it to a user are described. Products and services offered by various merchants are read using a merchant specific catalog and stored in a common format. Categories for such products and services are normalized and virtual categories are created using various product attributes. Visual creatives, termed as ad-templates are created to control the visual and interactive aspects of the ad, including ad-size, color, as well as product attributes that are displayed in the ad. Ad-templates may be constrained to specific products or product categories. A learning algorithm uses an adaptive sampling process to sample various products, product categories and ad-templates independently for different learning units such as individual users, groups of users determined by some demographics, individual web pages and groups of web pages grouped using various similarity criteria. The performance of the ad is measured using various learning statistics, such as the click-through-rate, conversion rate, etc. The learning algorithm uses the learning statistics to optimize the return for the advertiser by favoring the products or categories that perform better on one or more specified criteria.