Real-Time Advertising Forecasting From Historical Search Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing forecasting methods for web-based advertising campaigns suffer from inaccuracy, leading to inefficient energy consumption and carbon footprint, particularly in online advertising, which consumes significant energy and generates substantial CO2 emissions.
Innovation Solution
A computer-implemented method using Apache Spark for large-scale data processing to build and save an advertisements forecasting accuracy report, incorporating a directed acyclic graph (DAG) and a real-time analytics database like Apache Druid for accurate forecasting, enhancing the precision of ad performance metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional forecasting methods are used for web-based advertising campaigns, then the system is simpler to implement, but forecasting accuracy deteriorates leading to inefficient energy consumption and carbon footprint
Solution Approach 1:
The system performs preliminary forecasting using Apache Spark to predict ad performance metrics before campaigns execute. By pre-calculating expected impressions, clicks, and conversions based on historical data and campaign parameters, the system enables better resource allocation and reduces wasted energy on underperforming ads.
Solution Approach 2:
The system implements feedback loops where actual campaign performance data is continuously collected and compared against forecasts. This feedback is used to refine forecasting models, improving accuracy over time and enabling dynamic optimization of ad spending to reduce energy consumption from ineffective advertising.
2Measurement precision
If traditional forecasting methods are used, then implementation is simpler, but forecasting accuracy deteriorates
Solution Approach 1:
The forecasting system is segmented into modular components: data collection modules, processing modules using Apache Spark, storage modules, and analysis modules. This segmentation allows complex forecasting tasks to be broken down into manageable parts that can be developed, maintained, and scaled independently while achieving high accuracy.
Solution Approach 2:
Apache Spark serves as an intermediary computing engine between raw advertising data and forecasting results. It mediates the complex data processing requirements by providing a unified framework for distributed computation, handling everything from data ingestion to model execution without requiring custom infrastructure for each processing step.
3Productivity
If inaccurate forecasting is used, then fewer system resources are required, but more ad impressions are needed to compensate, increasing energy usage and bandwidth requirements
Solution Approach 1:
The system performs preliminary forecasting using Apache Spark to predict ad performance metrics before campaigns execute. By pre-calculating expected impressions, clicks, and conversions based on historical data and campaign parameters, the system enables better resource allocation and reduces wasted energy on underperforming ads.
Solution Approach 2:
The system replaces brute-force trial-and-error ad deployment with intelligent predictive modeling. Instead of randomly testing many ad impressions to determine effectiveness, machine learning models predict performance upfront, substituting computational intelligence for mechanical trial-and-error processes and significantly reducing wasted impressions.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
There is disclosed a computer-implemented method of forecasting a number of clicks and a number of impressions a line item will receive in a defined time period, the method including the steps of: (i) a forecasting service receiving a request from a portal for a forecast of the number of clicks and the number of impressions a line item will receive in a defined time period, the request including targeting criteria and placement data; (ii) the forecasting service translating the targeting criteria and the placement data into an index querying syntax; (iii) the forecasting service using the index querying syntax to request inventory via a service; (iv) the service making a request corresponding to the index querying syntax for a time series of historical searches to a database system; (v) in response to the request, the service receiving the time series of historical searches from the database system; (vi) the service providing inventory to the forecasting service corresponding to the defined time period, based on the received time series of historical searches; (vii) the forecasting service using the returned inventory in a request to a real-time service which makes a forecast of the number of clicks and the number of impressions the line item will receive in the defined time period, in which the forecasting service receives the response from the real-time service in real time, e.g. in less than one second; (viii) the forecasting service using the received forecast of the number of clicks and the number of impressions the line item will receive in the defined time period, to derive a click-through-rate (CTR) for the line item; (ix) the forecasting service returning the forecast of the number of clicks and the number of impressions the line item will receive in the defined time period, and the click-through-rate (CTR) in the defined time period, to the portal. Related methods and systems are disclosed.