Adjusting Ad Auction Rankings via Predicted User Dislike Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies struggle to effectively adjust the ranking and pricing of online content items in real-time auctions based on user feedback, leading to inefficiencies in content selection and revenue optimization.
Innovation Solution
A data processing system that uses machine learning models trained with historical feedback signals to predict user response to content items, adjusting their ranking and pricing in real-time auctions based on predicted levels of dislike or annoyance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If traditional real-time auctions rank content items based on bid price and predicted click-through rate, then revenue maximization is achieved, but user experience deteriorates due to display of disliked content
Solution Approach 1:
The system performs preliminary actions by training machine learning models offline using historical feedback signals before real-time auction execution. The models are pre-trained to predict user dislike signals, enabling the system to anticipate user reactions and adjust auction rankings proactively rather than reactively, thus preventing poor user experience before it occurs
Solution Approach 2:
The system implements a feedback mechanism where user responses to content items (such as advertisements) are collected as feedback signals. These signals are used to train machine learning models that predict user dislike signals, which then feed back into the auction ranking system to adjust content item rankings, creating a closed-loop system that continuously improves both revenue and user experience
2Measurement precision
If machine learning models are trained with extensive historical feedback signals to improve prediction accuracy, then content selection quality improves, but system complexity increases
Solution Approach 1:
The system segments the complex machine learning task into distinct components: offline model training phase and online prediction phase. The offline training uses historical feedback signals to build prediction models, while the online phase simply applies these pre-trained models to new content items. This segmentation allows complex training to occur separately from time-critical real-time operations, reducing overall system complexity
Solution Approach 2:
The patent introduces an intermediary machine learning model that acts as a mediator between raw feedback signals and auction ranking decisions. Instead of directly processing complex feedback data during real-time auctions, the system uses the pre-trained model as an intermediary to translate feedback signals into predictive dislike scores, simplifying the real-time decision-making process
3Object-affected harmful factors
If the system adjusts auction rankings based on predicted dislike signals, then user experience improves, but processing time increases
Solution Approach 1:
The system performs the computationally intensive model training in advance during an offline phase, before real-time auction execution. This preliminary action ensures that when real-time auctions occur, the models are already trained and ready to provide predictions quickly, thus improving user experience without adding processing time to the critical real-time path
Solution Approach 2:
The system maintains continuous improvement of prediction models through ongoing training with new feedback signals while preserving the ability to perform real-time predictions. The offline training can continue in the background without interrupting real-time auction operations, ensuring both user experience improvement and timely content delivery
Data Source
AI summary
A method and system for adjusting ads auction using predicted user responses to an in-ad survey is provided. The method includes (1) providing a content item associated with an actionable object, which when selected, causes a computing device to present a plurality of interactive elements each corresponding to a different one of a plurality of reasons for restricting the content item; (2) receiving, from the computing device, data indicating a particular reason, of the plurality of reasons, for restricting the content item, and the particular reason corresponding to a particular interactive element, of the plurality of interactive elements, that was selected by the user; and (3) updating, using the received data, a content selection model for selecting content items, wherein the content selection model is associated with the user.