Browse Prior Art Database

Method and System for Using Approximate Event Detection to Make Remarkable Improvements in Delivery Guarantees

IP.com Disclosure Number: IPCOM000249076D
Publication Date: 2017-Feb-01
Document File: 5 page(s) / 92K

Publishing Venue

The IP.com Prior Art Database

Related People

Kaushik Sathupadi: INVENTOR [+3]

Abstract

A method and system is disclosed for detecting and approximating event occurrences with confidence intervals within a demand duration for generating and using dynamic forecast trends based on these confidence intervals. The method and system uses existing demand durations to generate dynamic forecast trends instead of statically forecasting for individual dates by generating confidence intervals for events and guarantees an increased impression count with an increase in demand duration.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 43% of the total text.

Method and System for Using Approximate Event Detection to Make Remarkable Improvements in Delivery Guarantees

Abstract

A method and system is disclosed for detecting and approximating event occurrences with confidence intervals within a demand duration for generating and using dynamic forecast trends based on these confidence intervals.  The method and system uses existing demand durations to generate dynamic forecast trends instead of statically forecasting for individual dates by generating confidence intervals for events and guarantees an increased impression count with an increase in demand duration.

Description

Disclosed is a method and system for detecting and approximating event occurrences with confidence intervals within a demand duration for generating and using dynamic forecast trends based on these confidence intervals.

The method and system uses existing demand durations to generate dynamic forecast trends instead of statically forecasting for individual dates by generating confidence intervals for events and guarantees an increased impression count with an increase in demand duration.

The method and system implements an algorithm that is based on Dynamic Forecast Trends (DFTs) utilized for advertising solutions.

The algorithm implemented by the method and system is both duration

aware and error aware.  Further, methods for calculating the expected revenue uplift of the algorithm are described as follows.

In accordance with an embodiment, the guaranteed booking software is a system consisting of bookings (B), supply samples (S) and trend nodes (T).  A demand node Di is connected to a supply sample Sj. via an edge Eij.  Every Sample Sj has a weight Wj and is connected to a single trend node Tk.

The total impression for any given query is constructed in a three-step process.  Firstly, a pseudo demand node is created with the given targeting attributes.  Secondly, the demand node is connected to an existing bipartite Graph (B, S). Finally, the total impressions are calculated as:

This calculation does not take into account optimization that includes re-allocating existing contracts with different durations to make more room for a

current contract as well as availability which is total minus the total that is booked out of all supply samples after optimization.

Further, since supply forecasting in guaranteed delivery is to maximize the overall profit, every guarantee made on a demand node di is associated with penalty amount Pi.

Also, the method and system considers two effects on duration namely the simple effect of the mathematics of error reduction with an increased number of samples and the effect of seasonality and events.  In the case of seasonal effects on supply, the goal is to magnify the problem of daily versus window based optimization based on considering the effects of “Event Shifting” and “Event Peaks”.

Event shifting is when an event predicted for a day has been shifted by one or more days.

In this case...