Demand Side Platform Winning Bid Price Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In real-time bidding (RTB) systems, Demand Side Platforms (DSPs) face challenges in efficiently predicting winning bid prices due to censored data, leading to inefficient use of computer hardware resources and suboptimal ad delivery, as they lack information on competitors' bids and can only observe winning prices for auctions they win.
Innovation Solution
A machine learning-based approach using a gradient boosting framework for censored regression is implemented, which estimates winning bid prices by combining historical win and loss data, calculating likelihood functions, and updating models to predict bid prices accurately, thereby optimizing hardware resource usage and ad delivery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If DSPs participate in real-time bidding auctions without accurate winning price prediction, then they can potentially win ad impressions, but they waste significant computer hardware resources on unsuccessful bids
Solution Approach 1:
The system performs preliminary action by training a machine learning model offline using historical bid data to predict winning prices before actual real-time bidding occurs. This pre-computed knowledge enables DSPs to make informed bidding decisions without wasting resources on unrealistic bids during live auctions.
Solution Approach 2:
The system implements feedback by continuously training the machine learning model with historical bid outcomes (both won and lost auctions). The model learns from past performance data to improve its winning price predictions, creating a closed-loop system that optimizes bidding strategy over time based on actual results.
2Measurement precision
If DSPs use traditional bidding strategies without censored data handling, then the bidding process is simple, but prediction accuracy of winning prices deteriorates due to lack of competitor bid information
Solution Approach 1:
The system converts the harmful effect of censored data (incomplete competitor bid information) into a benefit by treating lost bids as valuable training samples. The machine learning model learns from both won and lost bids, using the censoring pattern itself as a signal to infer competitor bidding behavior and improve prediction accuracy.
Solution Approach 2:
The system applies parameter changes by transforming the bid prediction problem into a survival analysis framework with censoring. Instead of treating missing competitor bid data as a problem, the model adjusts its parameters to handle censored observations, where the winning price is only partially observable (known to be above the losing bid but exact value unknown).
3Productivity
If DSPs bid on every ad impression opportunity, then they maximize potential ad delivery volume, but hardware resource usage increases significantly
Solution Approach 1:
The system applies partial action by having DSPs bid selectively rather than on every ad impression. The machine learning model predicts winning prices for potential bids, and the DSP only participates in auctions where the predicted winning price is within its budget constraints, achieving sufficient ad delivery volume while reducing unnecessary bidding activity.
4Reliability
If DSPs lack accurate winning price estimation, then bidding decisions are simpler to make, but profit margins decrease due to suboptimal bid pricing
Solution Approach 1:
The system introduces an intermediary machine learning model that acts as a mediator between historical bid data and bidding decisions. This model estimates winning prices by processing historical patterns and competitor behavior, providing reliable profit margin predictions without requiring complex real-time analysis during actual bidding.
Data Source
AI summary
The demand-side platform (DSP) is a technological ingredient that fits into the larger real-time-bidding (RTB) ecosystem. DSPs enable advertisers to purchase ad impressions from a wide range of ad slots, generally via a second-price auction mechanism. In this aspect, predicting the auction winning price notably enhances the decision for placing the right bid value to win the auction and helps with the advertiser's campaign planning and traffic reallocation between campaigns. This is a difficult task because the observed winning price distribution is biased due to censorship; the DSP only observes the win price in the case of winning the auction. For losing bids, the win price remains censored. In this invention, we generalize the winning price model to incorporate a gradient boosting framework adapted to learn from both observed and censored data. This yields a boost in predictive performance in comparison to classic linear censored regression.