In one embodiment, a viewing behavior subsystem computes estimated viewership among target audiences at a person-level based on a household viewing logs and person-level data. A viewing behavior subsystem distributes the household viewing logs across buckets based on the sizes of the households and sets of person-level characteristics that are associated with the persons within the households. The viewing behavior subsystem generates a model for viewing behaviors associated with different sets of person-level characteristics based on the buckets. Subsequently, the viewing behavior subsystem estimates viewership among a target audience that is associated with one or more of the sets of person-level characteristics based on the model and census data. Advantageously, by combining household viewing logs for a relatively large number of households with person-level data, the viewing behavior subsystem enables accurate estimations of viewership among target audiences that are distinguished by person-level characteristics.
BACKGROUND OF THE INVENTION Field of the Invention (1) Embodiments of the present invention relate generally to computer processing and, more specifically, to techniques for estimating person-level viewing behavior. Description of the Related Art (2) Estimating linear programming viewing behavior is an integral part of negotiating advertising contracts. “Linear programming viewing” refers to viewing programs as scheduled and aired by a broadcaster. As a general matter, as the number of persons included in a target audience for a particular product that likely will view a program increases, the more an advertiser of that product will pay to advertise during the program. Accordingly, to maximize advertising rates, broadcasters oftentimes provide “guarantees” with respect to the level of viewership among target audiences. For example, to maximize the amount an advertiser will pay for an advertisement for full-size trucks during a football game, a broadcaster could provide a guarantee of a minimum viewership of the game among males aged 18-49. To provide reliable guarantees to a wide assortment of advertisers, broadcasters need accurate estimates of historical viewership among a variety of target audiences. (3) In that regard, many broadcasters rely on active viewing logs to estimate viewership, where the active viewing logs are typically gathered from a viewing panel of participating households. For example, within each participating household, a specialized meter can track whic
1. A computer-implemented method for estimating viewership among target audiences, the method comprising: assigning a first household viewing log to a first bucket of one or more household viewing logs based on person-level data, wherein the first bucket is associated with a single-person household and a first set of one or more person-level characteristics; assigning a second household viewing log to a second bucket of one or more household viewing logs based on the person-level data, wherein the second bucket is associated with a multi-person household and the first set of one or more person-level characteristics; performing one or more machine learning operations to generate a model based on first viewing propensities associated with the first bucket and second viewing propensities associated with the second bucket, wherein the one or more machine learning operations comprise computing the second viewing propensities based on the first viewing propensities and co-viewing impacts associated with the second bucket; and computing a first viewership among a first target audience that is associated with the first set of one or more person-level characteristics based on the model and census data. ||
11. A computer-readable storage medium including instructions that, when executed by a processing unit, cause the processing unit to estimate viewership among target audiences by performing the steps of: assigning a first household viewing log to a first bucket of one or more household viewing logs based on person-level data, wherein the first bucket is associated with a single-person household and a first set of one or more person-level characteristics; assigning a second household viewing log to a second bucket of one or more household viewing logs based on the person-level data, wherein the second bucket is associated with a multi-person household and the first set of one or more person-level characteristics; performing one or more machine learning operations to generate a model based on first viewing propensities associated with the first bucket and second viewing propensities associated with the second bucket, wherein the one or more machine learning operations comprise computing the second viewing propensities based on the first viewing propensities and co-viewing impacts associated with the second bucket; and computing a first viewership among a first target audience that is associated with the first set of one or more person-level characteristics based on the model and census data. ||
20. A system comprising: a memory storing a viewing behavior engine; and a processor that is coupled to the memory and, when executing the viewing behavior engine, is configured to: assign a first household viewing log to a first bucket of one or more household viewing logs based on person-level data, wherein the first bucket is associated with a single-person household and a first set of one or more person-level characteristics, assign a second household viewing log to a second bucket of one or more household viewing logs based on the person-level data, wherein the second bucket is associated with a multi-person household and the first set of one or more person-level characteristics, perform one or more machine learning operations to generate a model based on first viewing propensities associated with the first bucket and second viewing propensities associated with the second bucket, wherein the one or more machine learning operations comprise computing the second viewing propensities based on the first viewing propensities and co-viewing impacts associated with the second bucket, and compute a first viewership among a first target audience that is associated with the first set of one or more person-level characteristics based on the model and census data.